THE SKINNY
on AI for Education
Issue 28, May 2026
Welcome to The Skinny on AI for Education newsletter. Discover the latest insights at the intersection of AI and education from Professor Rose Luckin and the EVR Team. From personalised learning to smart classrooms, we decode AI's impact on education. We analyse the news, track developments in AI technology, watch what is happening with regulation and policy and discuss what all of it means for Education. Stay informed, navigate responsibly, and shape the future of learning with The Skinny.
Headlines
The Human Provenance Premium
I have been on holiday the past few weeks and have been thinking a lot about Human Intelligence and the ways in which it is different and often superior to AI. More of that to come....
For this month’s Skinny on AI, I focus on a particular article and context that I found thought provoking, and of course the usual Skinny Scan of the news follows the editorial: both the 60 second and the full version.
The piece I focus on here is an essay by Alex Imas in his Substack: ‘Ghosts of Electricity’ about where the AI argument is going next. Imas is a behavioural economist at Chicago Booth. Shortly after the essay appeared, Google DeepMind announced him as their first Director of AGI Economics. He joins Ronnie Chatterji and Jason Furman at OpenAI, the group of ten economists Anthropic has convened, and the wider pattern Daron Acemoglu picked out in MIT Technology Review last month. The frontier AI labs are now investing in serious economic analysis of what advanced AI does to labour, wealth and institutions. Some of that is defensive, in response to public concern about jobs. But it is also a shift in how AI capability is being thought about inside the labs that build it.
In brief: the 60-second version
Human Provenance Matters
Imas's central question is one economists always ask. What becomes scarce? When AI makes a wide range of human production cheaper, society does not stop consuming. It reallocates spending. What it reallocates towards, the evidence says, is what Imas calls the relational sector: goods and services where the value is inseparable from the human providing them. He names them explicitly. Teachers. Nurses. Childcare workers. Trainers. Hospitality. Clergy. Guides.
The empirical finding inside the essay is interesting. Imas and Graelin Mandel ran a study on art prints, varying only the perceived involvement of AI. Human-made prints gained 44% in value from exclusivity (one copy versus many). AI-made prints gained only 21%. The mere involvement of AI made the print feel reproducible, and reproducibility erodes the value premium that comes from a thing being attached to a particular person, in a particular relationship, at a particular time.
Education is a relational good in exactly this sense. The expensive independent school is not selling content. It is selling the teacher who knows the child, the cohort that child sits alongside, and the institution whose reputation makes that cohort meaningful. Paul LeBlanc has been making a version of this case for some time. Imas now gives it an economic foundation.
The standard worry about AI in education is displacement: AI replaces the teacher. The Imas essay suggests a different and arguably more serious risk. AI does not have to replace the teacher to damage the relational good. It only has to leak into it. If AI involvement becomes visible in the marking, the feedback, the tutorial conversation, the pastoral note, the lesson planning, the human institutional offer begins to look more reproducible. The exclusivity premium that supports the unique value of human teachers begins to soften. The relational good is being quietly converted into a commodity one.
This sharpens rather than resolves the question of where AI sits in education. It tells us the relational core must remain visibly and integrally human. It does not tell us which tasks count as the relational core and which are a commodity overhead AI should absorb. That distinction is the work to do, and it is the challenge that AI literacy at institutional level needs to be increasingly answering.
The economic argument runs the other way too. If the human element is what AI cannot commodify, then teachers, nurses, carers and trainers are not the soft underbelly of the AI economy. They are where the AI economy will land, in spending and in employment. The training and education sector is not the residual category of the AI economy. It is the demand-rich centre of it.
For learning professionals: here is a question to take into your next AI strategy conversation: Where, in this institution, does human provenance matter? And where, by contrast, is automation a service to the relational core rather than a substitute for it? The answer will not be the same in any two institutions. But the question is.
The full 5-minute version
The economists arrive in the labs
Until very recently, the dominant voices inside the AI labs were technical: engineers, model researchers, safety teams. The shift this year has been that economists are now in the room as well, not as commentators but as employees. Imas at DeepMind. Chatterji and Furman at OpenAI. Ten economists convened at Anthropic. Acemoglu, writing in MIT Technology Review, reads the trend as the labs responding to public concern about jobs. That is part of it. The wider point is that the question of what AI does to the economy is now being treated as engineering-relevant, not just policy-relevant, by the institutions building the technology.
This matters for educators because the framework through which the labs talk about AI in society is becoming more economically literate. The categories that economists use (income elasticity, structural change, the labour share, Baumol's cost disease) are likely going to start appearing in the language the labs use about their own technology, and in the language policymakers use about regulating it. We should be ready for the conversation.
The Imas argument, put simply
The argument has three steps.
First, when a technology makes a category of production dramatically cheaper, society does not stop consuming what that category produces. It reallocates spending. People did not eat less when farming became efficient. They reallocated the money they used to spend on food to other things. Agriculture fell from about 40% of the US workforce in 1900 to less than 2% today, while food consumption rose. The classic structural-change pattern moved labour from farms to factories to services.
Second, what people reallocate towards as they get richer is not just more stuff. It is goods and services with higher income elasticity of demand: categories whose share of consumption rises faster than income rises. Comin, Lashkari and Mestieri estimate that this income effect accounts for over 75% of historical patterns of structural change. The price effect (the standard story that automated goods get cheap so people buy other things) accounts for only about a quarter. The bigger force is that as people get richer, they want fundamentally different things.
Third, what makes those goods income-elastic is not only their cost. It is their human provenance. Drawing on René Girard, Imas argues that mimetic desire (the idea that what we want is shaped by what others want, especially when others cannot have it) is comparative and therefore difficult to satiate. The exclusivity that makes something desirable is exactly what cannot be mass-produced. The hand on the page, the personal note on the cup, the teacher who knows the child by name: these resist the commodity form, and that resistance is what gives them their economic premium.
The Mandel finding
Imas ran an experiment with Graelin Mandel on art prints which is a nice consequential empirical piece for our sector.
Subjects were shown identical art prints, but the description was varied. In some conditions, the print was described as human-made. In others, as AI-made. In some conditions, only one copy existed; in others, the print was reproducible. The willingness-to-pay results are striking.
Human-made prints gained 44% in value when exclusive (one copy) compared with non-exclusive (many copies). AI-made prints gained only 21% under the same conditions. The mere involvement of AI made the print feel reproducible, even when subjects were explicitly told the number of copies was identical. AI involvement collapsed the exclusivity premium by more than half.
The implication is not that AI-generated work has no value. It is that AI involvement signals reproducibility, and reproducibility erodes the value premium that comes from a thing being attached to a particular person, in a particular relationship, at a particular time.
Education as a relational good
Education has always been, in part, a relational good. The teacher who knows the child, the cohort that child sits alongside, the institution whose reputation makes that cohort meaningful. Paul LeBlanc has been making a related case for some time, arguing that what schools sell is not content delivery but human formation. Imas gives that argument an economic foundation. The income elasticity of education, at least in its more selective and formative forms, is high precisely because the human provenance is integral.
This is why the most selective schools and universities continue to command premiums that AI content delivery does not erode. The thirty thousand pound a year school is not selling content. It is selling the teacher, the cohort, the institution, and the human formation those three produce together. A Khan Academy lesson, however well designed, does not displace what those schools are offering, because what those schools are offering is not the lesson.
And it is not only the most selective institutions. The same logic runs further down the price ladder than the elite-fees example suggests. The reason families and learners commit to a particular school, a particular sixth-form, a particular apprenticeship provider, a particular university, is in significant part the human texture of those institutions. The content overlaps. The relationships do not.
The risk we should be thinking about
The standard worry about AI in education is displacement: AI replaces the teacher. The Imas argument suggests a different and arguably more serious risk. AI does not have to replace the teacher to damage the relational good. It only has to leak into it.
The Mandel finding is the model for what happens then. When subjects believed AI was involved in producing the print, the exclusivity premium collapsed even though the print itself was identical. The signal of AI involvement was enough to convert a human-provenance good into something that felt reproducible.
For schools and universities, the equivalent failure mode is direct. The relational core of an institution (the formative relationships, the cohort experience, the human-led pastoral life, the marking, the feedback, the tutorial conversation, the lesson planning that shapes what the child encounters) begins to be visibly assisted, augmented, or generated by AI. At each of those points, AI involvement is rational on its own terms. It saves time. It smooths workloads. It supports tired teachers. But each of them sits inside the relational core that the institution is actually valued for.
If, in aggregate, AI involvement becomes visible in those places, the institutional offer begins to look more reproducible. The exclusivity premium begins to soften. Not because anyone has done anything wrong. Because the signal has changed. The relational good is being quietly converted into a commodity one.
This is not an argument against using AI in education. It is an argument that the place where AI sits inside an institution is a strategic question, not a operational one. The Imas essay tells us the relational core must remain visibly and integrally human. It does not tell us where the line is. The work to be done is to define, for any particular institution, what counts as the relational core (which AI should not touch except in service of) and what counts as commodity overhead (which AI absorbing is a service to the core, not a substitute for it).
What follows for policy and curriculum
The economic argument also runs in the opposite direction. If the human element is what AI cannot commodify, then the people who carry the human element in society (teachers, nurses, carers, trainers, mentors) are not the soft underbelly of the AI economy. They are where the AI economy will land, in spending and in employment. This makes two things true at the same time.
The pressure on existing professional structures is real, documented and accelerating. The News Skinny Scan covers this in detail: Standard Chartered's 8,000 cuts described by its CEO as 'lower-value human capital', Cloudflare's 20% reduction citing a 600% rise in AI usage, the NHS draft workforce plan proposing hundreds of thousands fewer staff, McKinsey's Project Acorn shifting partner pay toward equity, Big Four AI specialist postings now running at over twice the rate of audit postings. These moves are real.
And the long-run demand for the categories at the relational core (teaching, care, formation, hospitality) is, if Imas is right, structurally increasing. The training and education sector is not the residual category of the AI economy. It is the demand-rich centre of it. The implication for policy is that workforce planning that treats the relational professions as costs to be controlled has the economics backwards. They are the categories spending and employment are reallocating towards.
For curriculum design, the implication is sharper still. AI literacy that prepares learners for codifiable cognitive tasks is preparing them for the part of the labour market AI is most directly substituting. AI literacy that prepares learners to do work where their judgement, attention, memory, warmth or presence is part of the value is preparing them for the part of the labour market AI cannot enter. The Royal Society and TeacherTapp data this spring showed that the capabilities most teachers feel least confident teaching (evidence evaluation, model limitations, ethical reasoning) are precisely the capabilities that would equip learners for that part of the economy.
What to do with this
Two things, that I would offer to any reader to consider.
The first is to take the Imas framework into the AI strategy conversations in your school, university and training provider, and ask the question it makes possible. Where, in this institution, does human provenance matter? And where is automation a service to the relational core rather than a substitute for it? The answer will not be the same in any two institutions. But the question is the same, and the institutions that work it out first will have a clearer story to tell parents, students and their own staff than those that do not.
The second is to take the Imas-Mandel finding (the 44% to 21% collapse) into AI literacy teaching itself. It is a piece of empirical evidence that captures something subtle and important about how AI changes the value of human work. It belongs in the curriculum, and it belongs in the conversation about why work that has human provenance is not a luxury or a nostalgia but the part of the economy that grows as the rest gets cheaper.
The frontier labs are increasingly listening to economists. Education is one of the categories economists are now telling them matters most. We should be ready to make the case for what education actually is, in terms that an economist will recognise. Provenance. Formation. Exclusivity. The human being who is not just an input into the production process but part of the value.
Human provenance matters.
Sources: Alex Imas, 'What will be scarce? The economics of structural change and the post-commodity future', Ghosts of Electricity, 14 April 2026; Imas and Madarasz, working paper on mimetic preferences and willingness to pay; Imas and Mandel, working paper on AI involvement and the exclusivity premium; Diego Comin, Danial Lashkari and Marti Mestieri, 'Structural Change with Long-Run Income and Price Effects', Econometrica, 2021; Joachim Hubmer, 'The Race Between Preferences and Technology'; David Autor and Neil Thompson, working paper on AI, expertise and labour market outcomes; René Girard on mimetic desire; Daron Acemoglu, MIT Technology Review, April 2026, on the labs' turn to economic analysis; Paul LeBlanc, Learning Mate, on the human core of education in an AI age; Saffron Huang on positive AI-driven structural change; AI Landscape Tracker, May 2026 (Standard Chartered, Cloudflare, NHS workforce plan, McKinsey Project Acorn, Big Four hiring data).
The Skinny Scan
In brief: the 60-second version
A substantial piece of evidence on AI in the workplace this month is a measured gulf between what employers say AI is doing in their organisations and what workers experience. The Policy Institute at King's College London surveyed 4,500 UK respondents in April: 86% of employers report at least modest productivity gains from AI, against just 47% of AI-using workers who say it makes any difference to their performance. 69% of businesses are optimistic about AI's effect on job opportunities; only 35% of employees and 28% of the public agree.
There have also been more announcements about job cuts. Standard Chartered cut 8,000 jobs, with CEO Bill Winters describing them as 'lower-value human capital'. Cloudflare cut 20% of its workforce citing a 600% rise in AI usage in three months. The NHS draft workforce plan proposes hundreds of thousands fewer staff than the 2023 forecast. Big Four AI-specialist hiring now runs at over double the rate of audit hiring.
The Delegation Dilemma acquired a new name :). BCG researchers (n=1,488) introduced 'AI Brain Fry': mental fatigue from excessive use or oversight of AI tools, distinct from burnout. METR's new survey of 350 knowledge workers found that when respondents are asked about value rather than speed, self-reported AI gains drop from 3x to between 1.6x and 2x. The cognitively cheap parts of work get cheaper. The expensive parts get harder.
The cybersecurity picture is systemic. The IMF warned that AI models now 'elevate cyber risk to a potential macro-financial shock', citing Anthropic's Mythos finding vulnerabilities in every major operating system and browser. Trump postponed the AI executive order he had been scheduled to sign on 21 May. Bug bounty programmes are buckling under AI-generated submissions (HackerOne +76% year-on-year, with only 25% legitimate). And bias and accuracy also remain a problem. For example, Forum AI tested four major chatbots on more than 3,100 questions and found that 90% of election answers failed on accuracy, bias or source.
Public sentiment about AI is moving in the opposite direction to deployment. Pew: 50% of US adults are more concerned than excited about AI; only 10% are primarily excited. Gallup and the Walton Family Foundation: Gen Z AI excitement fell 14 percentage points in a year, with anger now at 31% and anxiety at 42%. The cohort that should be most native to AI is showing rising resistance, not rising adoption. 82% of Americans back the White House safety testing that Trump postponed.
Voices I have found worth attending to this month. Andrew Likierman (formerly Dean, London Business School) and Richard Ovenden (Bodleian Libraries) both argued in the FT that the qualities AI cannot supply, namely judgement, trusted stewardship, and decision-making under uncertainty, are precisely what professional education and public institutions need to invest in now. Business schools (INSEAD, HEC, Essec, UPF) are already there: UPF's Giulio Toscani: 'What AI systems cannot do is sign. But someone must.'
For training and education professionals: the 47/86 gap I highlight in the first paragraph here is not a gap in the technology. It is a gap in how the people doing the work are being prepared to use it. That is what training is for. I don’t think there has been a more actionable month in this debate for two years.
The full Skinny Scan
Cross-cutting patterns from this month's evidence, organised around the threads that my wider tracker is now following. Every item is filtered through three lenses where relevant. AI as a tool: what practitioners can use. AI as a catalyst: what AI demands of human intelligence in response. AI as a subject: what people need to understand about how these systems behave and why.
This was a heavy month. Five threads dominate, and a sixth is emerging fast enough to name. The most striking single number this year landed in a UK survey published this month: a measured gulf between what employers say AI is doing in their organisations and what workers experience. The disconnect, not the technology, is the actionable problem.
Thread 1: The Training Gap and the 47/86 disconnect
Central lens: Tool, Catalyst
The Policy Institute at King's College London published 'AI and the Future of Work' this month: a UK survey of 4,500 respondents across four groups (general public, young people, university students, employers), conducted 16 to 29 April 2026. Two findings dominate. 69% of businesses are optimistic about AI's effect on job opportunities, against 35% of employees and 28% of the general public. And, more striking, 47% of workers who use AI say it makes no real difference to their performance, against 86% of employers who report at least modest productivity improvement.
The gap is not a measurement error. Workers and managers are looking at the same technology and seeing different things. For training and L&D, this should be one of the most actionable findings of the year. If half of the people doing the work do not believe AI is helping them, no top-down adoption mandate will close that gap. The disconnect signals where deployment is failing: AI tools are being treated as productivity instruments without the workflow redesign or skill development that would let workers experience the gains.
Further structural moves this month... Standard Chartered announced 8,000 job cuts, with chief executive Bill Winters describing the change as 'replacing, in some cases, lower-value human capital' with AI. Cloudflare cut 20% of its workforce, around 1,100 staff, citing a 600% rise in AI usage in three months. The NHS draft workforce plan, seen by the FT, proposes using AI to avoid hundreds of thousands of forecast hires, warning that the existing recruitment path is 'a path to financial ruin'. McKinsey shifted partner pay towards equity in Project Acorn, driven partly by AI doing the work juniors traditionally billed by the hour. And FT analysis showed Big Four AI-specialist job postings now make up about 7% of all postings, against under 3% for audit; the AI figure has tripled since 2022.
Some institutions and policymakers are now investing in the gap. Multiverse, Euan Blair's apprenticeship-focused training company, raised $70mn at a $2.1bn valuation, with the explicit goal of training frontline workers rather than executives. Lord Gus O'Donnell, formerly head of the UK civil service, used his inaugural Downing Battcock Institute lecture to argue that AI's winners must fund retraining for AI's losers through tax. California Governor Gavin Newsom went further, returning to the idea of 'universal basic capital' as a response to AI-driven economic disruption: a mechanism that would give workers a financial stake in the productivity gains generated by AI systems, alongside more conventional retraining. The Recruiters' 'AI doom loop' reported by the FT this month adds the other half of the picture: companies now overwhelmed by AI-assisted applications they cannot tell apart, while candidates use AI to apply to roles they cannot meaningfully prepare for. The gap between adoption and training investment is now being empirically documented. The 47/86 number is the cleanest single articulation of why training, not technology, is the bottleneck.
Thread 2: The Delegation Dilemma
Central lens: Catalyst
The thread acquires a new dimension this month, and the persuasion mechanism that drives it sharpens further. Boston Consulting Group researchers (Bedard et al., n=1,488) introduce 'AI Brain Fry' as a construct distinct from burnout: mental fatigue from excessive use or oversight of AI tools beyond cognitive capacity. Symptoms include a 'buzzing' feeling, difficulty focusing, slower decision-making, decision fatigue, and intention to quit. Worst affected: software engineering and marketing (26% of marketers report it). Bedard's diagnosis is precise. When workers outsource repetitive tasks but keep judgement and creativity for themselves, they 'keep the most cognitively challenging part of the job with the human' while dealing with the increased throughput AI enables.
This sits alongside three existing dimensions previously discussed : cognitive debt (Kosmyna et al.), moral cost (Köbis et al.) and social-relational cost (Cheng et al.). The cumulative picture is that AI shifts cognitive load rather than reducing it. The cognitively cheap parts of work get cheaper; the cognitively expensive parts (judgement, framing, synthesis, oversight) get more intense.
A separate finding sharpens the persuasion side of the problem, and connects directly to the Sycophancy Trap. Tim Harford's column cited a Nature paper finding that LLMs trained to be warm and friendly produce dramatically less accurate answers: 'promoting conspiracy theories, providing inaccurate factual information and offering incorrect medical advice'. The training that makes AI feel helpful is the training that erodes accuracy. Harford frames LLMs as confidence tricksters: 'the sycophantic AI not only produces mistakes, it persuades us to believe them'.
And the productivity story is being reframed at the level of the source data. METR, the AI research non-profit, followed its earlier finding (that engineers felt AI made them 20% faster but precise measurement showed they were 20% slower) with a survey of 350 technical knowledge workers using value-based rather than speed-based questions. When asked about value rather than speed, self-reported gains fall from around 3x to between 1.6x and 2x. The less code-heavy the work, the smaller the boost. METR treats the figures as upper bounds.
The behavioural response to adoption mandates belongs here as well. FT reporting found Amazon employees using an in-house tool called MeshClaw to automate unnecessary tasks in order to inflate their AI usage scores. The company had set targets for over 80% of developers to use AI each week and had begun tracking token consumption on internal leaderboards. Meta employees are doing the same; the practice has earned a name, 'tokenmaxxing'. When AI usage is mandated, workers respond by manufacturing usage rather than learning to use AI well. The signal is consistent with the King's College finding: adoption mandates produce performance, not capability.
This same pattern has a more troubling face. The Information reported this month that Microsoft, Meta and xAI are increasingly using internal employee activity as proprietary training data: GitHub Copilot and VSCode telemetry at Microsoft, mouse movement and browser interaction tracking through Meta's internal 'Model Capability Initiative', and (more strikingly) cash offers from xAI to employees in exchange for their tax-return data to train Grok. The line between mandated AI use and surveillance of the workforce that uses it is now visibly thin. For AI literacy and for the ethics curriculum, this is a worked example of how the data economy quietly extends into the relationship between the worker and the employer.
Thread 3: The Hot Mess Reframe
Central lens: Subject
This is about the argument that AI risk is correlated failure at speed, not malign coherence. The IMF published a blog post on 7 May warning that the latest AI models 'elevate cyber risk to a potential macro-financial shock'. The Fund's exact framing is worth exploring for teaching: cyber risk is 'increasingly about correlated failures that could disrupt financial intermediation, payments and confidence at the systemic level'. The specific trigger: Anthropic's Mythos model, released to 40 mostly US-based organisations, has 'found thousands of high-severity vulnerabilities, including some in every major operating system and web browser'. Many non-US banks have been left without access to patches.
In the same period, Trump postponed the AI executive order he had been scheduled to sign on 21 May. Reporting suggests he was alarmed by the vulnerabilities Mythos exposed in the banking system. An Institute for Family Studies poll found 82% of Americans back White House safety testing for advanced models; the postponement was a federal regulatory step away from public sentiment, not towards it.
Bug bounty programmes are buckling under AI-generated submissions. HackerOne reports submissions up 76% in the year to March, with the legitimate share holding steady at 25%. Curl suspended its programme in January, citing an 'explosion in AI slop reports'. Nextcloud suspended in April. Cyber-security firms reported a 'third cohort' of 'experienced AI builders' who have developed automated end-to-end scanning and submission systems creating 'absolute carnage'.
Two further developments this month belong in this thread. Forum AI tested four major chatbots (ChatGPT, Gemini, Claude and Grok) against more than 3,100 questions on politics, healthcare and foreign affairs. Collective answers on elections failed on accuracy, bias or source selection 90% of the time. 36% contained at least one factual error (Grok: 52%). Perhaps most concerning: 35% of foreign-policy responses cited state-controlled sources including Russia's RT and China's Global Times and CGTN. And on the AI-generated political imagery side, FT analysis of Trump's Truth Social posts found AI-generated images surged sevenfold in May. The deepfake expert Henry Ajder called it a 'systemic embrace' of 'slopaganda' as a means of communication.
AI capability scales faster than human review, and the systems designed to evaluate AI outputs (peer review, bug bounty triage, journalistic fact-checking, regulatory oversight) are collapsing under the volume. The relevant preparation for AI literacy that is needed is not for a coherent adversary. It is for systems that produce confident output faster than human or institutional verification can keep pace.
Thread 4: Assessment in Crisis
Central lens: Tool, Subject
A Verge investigation on 16 May documented that AI-generated research papers are overwhelming peer review. Peter Degen of the University of Zurich Center for Reproducible Science and Research Synthesis traced a citation surge of a 2017 statistical paper to Chinese paper mills using Bilibili tutorials to teach how to produce publishable research in under two hours using AI writing assistance. Papers use public datasets such as the Global Burden of Disease study to generate predictions about disease X in population Y. Degen's verdict: 'AI currently holds the potential to bring down the publishing system as we know it.'
The dynamic is the same one affecting open-ended student assessment, but at the level of the most quality-controlled assessment process in society. Earlier paper mills produced flagrant errors that publishers could screen out. The new generation produces work that is wrong but plausible, and the better AI gets, the harder the detection becomes.
This is an important assessment story. Not because peer review is the relevant assessment context for most learners, but because if the most rigorous assessment process in society is buckling, then the dynamics affecting school and university assessment are not solvable through detection technology. Assessment redesign is the only structural lever, and the case for that argument is now stronger than at any previous point.
A separate dimension of the same story... A coalition of publishers including Hachette, Macmillan and (notably for the academic sector) Elsevier sued Meta and Mark Zuckerberg, alleging that the company illegally used copyrighted books and journal articles to train its Llama models, including by accessing pirated material and removing attribution data. If the case proceeds on those terms, the legal precedent for what counts as legitimate AI training data will be set in part by an academic publisher's lawsuit against an AI lab. For HE leaders, the implications for library and licensing policy, for institutional repositories, and for the long-running open access debate are direct.
Thread 5: The Apprenticeship Gap
Central lens: Catalyst
The structural moves in professional services this month put fresh evidence in this gap. The Big Four data (AI specialist postings up from under 2% to around 7% in three years, against audit at under 3%) shows the entry-level pipeline reshaping at the firms that built the modern professional-services apprenticeship. McKinsey's Project Acorn shifts the partner pay structure because AI is doing work that juniors traditionally billed by the hour. The NHS workforce plan, which proposes hundreds of thousands fewer staff than the 2023 forecast, applies similar logic to clinical training: the entry-level pipeline through which clinicians acquire tacit knowledge is being squeezed by fiscal as well as technological pressure.
With respect to the equity dimension: Brookings estimates that of the roughly 6mn US workers most exposed to AI-driven displacement and least equipped to navigate it, more than 85% are women. The International Labour Organisation found that roles at the highest risk of AI-driven task automation account for 9.6% of female employment in higher-income countries, nearly triple men's share. A Harvard study found women using AI at a 25% lower rate than men. The US gender pay gap has expanded in the past two years after decades of progress. The disappearance of junior tacit-knowledge-building roles is landing disproportionately on women.
New thread proposed this month: The Adoption-Sentiment Inversion
Central lens: Subject, Catalyst
AI adoption is accelerating while public, worker and Gen Z sentiment is moving in the opposite direction. This is distinct from the KPMG Trust Paradox, which concerned declining trust in the specific tool a user uses. It is about how the population at large feels about AI as a phenomenon, even as deployment accelerates.
The Pew Research Center's five-year surveys find that 50% of US adults are more concerned than excited about AI in daily life. Only 10% are primarily excited. 18% of workers say their job will be 'very or somewhat likely' eliminated by AI within five years, up from 15% in mid-2025.
The Gallup and Walton Family Foundation data is sharper still. Gen Z AI excitement has fallen 14 percentage points in a year. Anger has risen to 31%; anxiety stands at 42%; excitement is now 22% and hopefulness 18%. Gen Z weekly genAI use is essentially unchanged from 2025, at just over half, at the same time as broader worker access rose 50% and the share of work hours using genAI climbed from 4.1% to 5.7%. The cohort that should be most native to AI is showing rising resistance, not rising adoption.
The political dimension... The 82% support for White House safety testing recorded by the Institute for Family Studies poll became visible the day Trump postponed the order that would have introduced it. Edward Luce's FT column on 12 May, 'Why Americans dread AI', captured the same gap from the other side: public scepticism about AI's societal impact is widening at exactly the moment Silicon Valley and Washington are accelerating investment. Declining consent for AI deployment now coexists with accelerating institutional adoption.
For education and training, the implications are direct. Course design assumptions about students wanting more AI in teaching should be tested against actual cohort views. HE recruitment positioning needs to explore and respond to declining enthusiasm in their cohorts. And the question of who AI is being deployed for, on whose terms, is now a question students themselves are asking with increasing scepticism.
What good practice looks like, and where institutional voices are landing
Business schools showed what a maturing in their AI-era curricula. The FT executive education special report this month profiled INSEAD, HEC Paris, Essec, UPF Barcelona, Polimi, Esade and Essca all moving beyond AI basics to teaching judgement under AI. INSEAD's Sameer Hasija frames it as 'humans thinking about the problem and pushing forward with their own judgement, while augmenting their information with AI'. HEC's AItelier platform guides executives through translating business problems into AI use cases; David Restrepo describes the output as 'collective refinement' rather than automated answers. Essec's Thomas Huber teaches 'AI-fluent' leaders to 'push back' on AI recommendations.
Perhaps the sharpest formulation comes from UPF Barcelona's Giulio Toscani: 'What AI systems cannot do is sign. But someone must.' UPF workshops require executives to sign off on AI-driven decisions on loan denials, job rejections, and medical resource allocation. The teaching is parallel to the Learning Mastery argument: the meta-skills of judgement under AI (when to trust, when to question, when to override) are what professional education needs to develop. The benchmark exists; the question for mainstream university and vocational provision is how to get there.
Andrew Likierman, formerly Dean of London Business School, made the corresponding argument in the FT on 17 May. As AI automates more routine analytical and operational work, the strategic value of human judgement increases rather than decreases. The qualities Likierman names (awareness, trust, emotional intelligence, decision-making under uncertainty) are not soft alternatives to AI capability; they are what determines whether AI capability produces value. For training and L&D, this is independent reinforcement of the Catalyst argument from one of the senior voices in management education.
Richard Ovenden, Director of Bodleian Libraries, used a separate FT piece on 18 May to argue that libraries should be central to the AI era, not peripheral to it. His framing is that libraries are the trusted stewards of knowledge, data and digital infrastructure that AI training and AI evaluation both require, and that years of fragmented UK library policy and underinvestment have left a strategic asset undefended at exactly the moment it matters most. For the broader argument that public institutions need to lead a different kind of AI conversation, this is a useful and timely addition from one of the most respected voices in the field.
Economic and infrastructure developments
The financial picture continues to evolve in ways that will affect what AI services institutions can buy and at what cost. Anthropic is on track for its first profitable quarter ($559mn operating profit on $10.9bn Q2 revenue), at a $900bn valuation with a $30bn round closing, and has reportedly passed OpenAI on the Ramp AI Index for enterprise adoption (34.4% vs 32.3%). OpenAI is preparing an IPO filing for as soon as September, at a potential valuation above $1tn. NextEra and Dominion announced a $420bn all-stock merger to control 'data centre alley' near Washington. Nvidia has committed roughly $90bn across 145+ AI companies in the past 16 months, embedding itself across model labs, cloud providers and infrastructure groups in a way analysts now describe as systemically important. Big Tech 2026 AI capex is estimated at $725bn, with foreign-currency debt now around 30% of hyperscaler borrowing. Alphabet sold a 100-year sterling bond in February. The cost of AI is increasingly visible in household electricity bills (Virginia power costs are up 12% since February 2025) before its benefits are visible in worker productivity surveys.
The same month produced the clearer scepticism about whether the maths works. The FT's 'Is an AI spending plateau on the horizon?' (14 May) found most surveyed financiers expecting the current pace of capex to remain sustainable for only one to three more years. CB Insights reported that private AI companies raised over $226bn globally in Q1 2026, surpassing the total raised in all of 2025. Joachim Klement's 19 May FT opinion piece compared the current build-out unfavourably with the dotcom bubble and questioned whether hyperscalers can generate enough future revenue to justify the spend. Nvidia beat expectations again and saw shares fall. The macro question, in short, is whether the revenue ever catches up with the infrastructure, and the consensus that there is no upper bound is starting to crack yet.
Meanwhile the agentic consumer push has become ever more coherent. Meta is developing an autonomous personal assistant powered by its Muse Spark model and explicitly framed as a Meta answer to OpenClaw. Google announced AI agents inside its search engine alongside renewed smart-glasses ambitions in partnership with Samsung, Warby Parker and Gentle Monster. OpenAI and Plaid are partnering to give ChatGPT users personalised financial advice. The next deployment frontier is consumer agents that hold and use personal data, including financial and health information. The AI literacy curriculum will need to keep pace with what these systems do, what they retain, and what they delegate to whom.
DeepSeek is nearing a $45bn valuation with China's state-backed Big Fund leading the round. DeepSeek V4 is optimised for Huawei's Ascend 950PR chips. Nvidia's Jensen Huang said openly that the day DeepSeek runs best on Huawei would be 'a horrible outcome for our nation'. The global AI ecosystem may be developing two parallel stacks rather than one converging market, with implications for what AI services will be available in which jurisdictions over the next several years.
The Skinny is published monthly for education and training professionals. The evidence tracker is available on request.
Sources for this issue: Policy Institute at King's College London (Duffy et al.), 'AI and the Future of Work', May 2026; Boston Consulting Group (Bedard et al.), 'AI Brain Fry', May 2026; METR (2026), Subjective Value of AI to Knowledge Workers; Nature (2026) on warm-friendly LLM training, cited in Tim Harford, 'Conned by a chatbot', FT, 5 May 2026; IMF blog on AI-enabled cyber risk, 7 May 2026; Forum AI chatbot reliability study, May 2026; The Verge (Dzieza), 16 May 2026, on AI-generated research papers and peer review; FT (Murphy), 5 May 2026, on Meta copyright lawsuit by Hachette, Macmillan and Elsevier; FT analysis of Trump's Truth Social posts, 19 May 2026; FT executive education special report, 17 May 2026; Andrew Likierman, FT, 17 May 2026; Richard Ovenden, FT, 18 May 2026; Edward Luce, 'Why Americans dread AI', FT, 12 May 2026; Andrew Ross Sorkin, NYT DealBook, 22 May 2026, on Newsom's universal basic capital proposal; FT (Telford), 9 May 2026, on women and AI displacement; FT (Smith), 14 May 2026, on Multiverse; FT (Alim), 18 May 2026, on Standard Chartered; The Information (Dean), 8 May 2026, on Cloudflare; FT (Smyth), 18 May 2026, on the NHS workforce plan; FT (Murray and Kissin), 18 May 2026, on the Big Four; FT (Kissin and Foley), 14 May 2026, on McKinsey; FT (Smyth), 6 May 2026, on Lord O'Donnell; FT (Rosner-Uddin), 11 May 2026, on Amazon MeshClaw; The Information, 19 May 2026, on Microsoft, Meta and xAI training data from employees; FT (Balakrishnan), 3 May 2026, on recruiters' AI doom loop; FT (Muir), 14 May 2026, on the AI spending plateau; Joachim Klement, FT, 19 May 2026, on the maths of the AI boom; CB Insights (Sanwal), 'AIming high', April 2026; FT (McMorrow and Acton), 19 May 2026, on Nvidia dealmaking; Ramp AI Index, via DeepLearning.AI, 19 May 2026, on Anthropic business adoption; FT (Murphy), 5 May 2026, on Meta's agentic assistant; FT (Morris and Criddle), 19 May 2026, on Google smart glasses and search agents; Stanford HAI AI Index Report 2026 Part II (figures via newsletter summary, full report to be verified); Pew Research Center; Gallup and Walton Family Foundation 2026 Gen Z AI survey; Institute for Family Studies poll, May 2026.
The Skinny News Items:
AI in Education
Euan Blair’s Multiverse hits $2.1bn valuation in AI workforce training push
14 May 2026 | Kieran Smith, Financial Times
Euan Blair’s education technology company Multiverse has raised $70mn at a $2.1bn valuation as it expands into AI workforce training and further European growth. The company, which trains and places apprentices, says AI is both a threat and an opportunity for workers. Blair argues that the technology could either destroy jobs or enhance productivity depending on how effectively organisations train employees to use it.
What you need to know: AI disruption is creating demand for workforce retraining companies. The funding round shows that investors see reskilling and applied AI training as a major growth market as employers try to adapt existing staff to new tools.
Original link: https://www.ft.com/content/ec5764f0-b783-48ac-9f16-a0ecb9da9636
Business schools move beyond the basics to teach collaboration with AI
17 May 2026 | Ian Wylie, Financial Times
Business schools are shifting executive education from basic AI literacy toward teaching leaders how to collaborate with AI in decision-making. The article highlights Uniqa Insurance Hungary’s NiQA system, which can analyse household insurance claims, interpret documents, calculate losses and authorise payments up to a defined threshold without human intervention. Schools such as Corvinus-SEED and Insead are using these examples to teach executives when to trust AI, when to question it and how to redesign decision rights between humans and machines.
What you need to know: AI education for leaders is moving from tool use to organisational redesign. The key management challenge is no longer simply understanding AI, but deciding how authority, accountability and oversight should be shared between humans and AI systems.
Original link: https://www.ft.com/content/b90ab9aa-a5cc-4f2d-8f28-30fbec4b3984
AI Regulation and Legal Issues
Opinion Today: A.I. as a national security threat
4 May 2026 | The New York Times
The New York Times Opinion Today newsletter focuses on the argument that artificial intelligence should increasingly be treated as a national security issue. The newsletter frames AI not only as a consumer or productivity technology, but as a system with potential implications for cyber defence, geopolitical competition and public safety. It reflects a broader shift in public debate toward whether the most powerful AI models should be governed through security-focused institutions and oversight mechanisms.
What you need to know: AI policy is moving from a technology-regulation debate toward a national-security debate. This matters because once AI is framed as a security threat, governments are more likely to pursue pre-release evaluations, export controls, compute monitoring and closer partnerships with major labs.
Original link: https://www.nytimes.com/newsletters/opinion-today
DealBook: Google’s White House A.I. meeting
4 May 2026 | Andrew Ross Sorkin, The New York Times
DealBook reports that Alphabet chief executive Sundar Pichai met senior Trump administration officials at the White House to discuss cybersecurity threats and the government’s access to AI processing capacity. The meeting followed concerns triggered by Anthropic’s restricted Claude Mythos Preview model, which reportedly raised fears in Washington about whether the US government had enough compute to maintain its own cyber defences. The newsletter also notes bullish sentiment at the Milken Institute Global Conference, where few finance leaders appeared to believe AI was in a bubble.
What you need to know: AI compute is becoming a national-security concern, not just a private-sector infrastructure issue. The meeting shows how advanced cyber-capable models are forcing governments to think about access to compute, defensive capacity and partnerships with major AI companies.
Original link: https://www.nytimes.com/newsletters/dealbook
Meta and Zuckerberg sued by publishers over ‘massive’ copyright infringement
5 May 2026 | Hannah Murphy, Financial Times
Meta and chief executive Mark Zuckerberg are facing a lawsuit from a coalition of major publishers, including Hachette, Macmillan and Elsevier, over claims the company illegally used copyrighted books and journal articles to train its Llama AI models. The lawsuit alleges Meta accessed pirated materials and unauthorised internet scrapes at massive scale, while also removing attribution data to conceal the sources of its training datasets. The case adds to mounting legal pressure on AI companies over the use of copyrighted material without consent or compensation.
What you need to know: Copyright disputes are becoming one of the defining legal risks facing generative AI companies, with the outcome likely to shape future AI training practices and licensing models.
Original link: https://www.ft.com/content/079ef5b2-5c68-435f-9f67-e02bd9073610
DealBook: Trump’s rethink on A.I.?
5 May 2026 | Andrew Ross Sorkin, The New York Times
DealBook examines whether the Trump administration is moving away from its earlier laissez-faire approach to artificial intelligence. The White House was reportedly considering an executive order that could create a working group of government officials and industry leaders to discuss oversight procedures, including a possible review process for new AI models before release. The newsletter frames the debate around whether AI models need something like an FDA-style approval process, while noting likely pushback from parts of the technology industry.
What you need to know: The article captures a possible turning point in US AI governance. Even an administration focused on maintaining America’s lead over China is being forced to consider pre-release oversight as frontier models become more powerful.
Original link: https://www.nytimes.com/newsletters/dealbook
Apple reaches $250mn settlement over delayed ‘AI Siri’
5 May 2026 | Michael Acton, Financial Times
Apple has agreed to pay $250mn to settle a false advertising lawsuit over AI features it promoted in 2024 but had not yet released. The lawsuit, brought on behalf of US iPhone buyers, claimed Apple misled customers by advertising a personalised AI-enhanced Siri that was still unavailable. The settlement comes as Apple prepares to unveil its delayed AI Siri and continues to face pressure after falling behind rivals in generative AI.
What you need to know: AI product promises are becoming legally risky when companies advertise capabilities before they are ready. The case shows how consumer-facing AI hype can create reputational and financial consequences if delivery lags behind marketing.
Original link: https://www.ft.com/content/f2c6a27e-8ed9-487c-9384-7e0f0dca3061
DealBook: Who invited the bot?
9 May 2026 | Andrew Ross Sorkin and Sarah Kessler, The New York Times
DealBook explores the legal risks created by AI note-takers and searchable meeting records. Lawyers are increasingly concerned that AI assistants in virtual meetings could create discoverable records of conversations that were previously informal, off-the-record or protected by privilege. The newsletter describes how some lawyers now ask participants to remove AI note-takers before meetings begin, reflecting wider anxiety about privacy, litigation exposure and the consequences of turning everyday work into searchable data.
What you need to know: AI is changing corporate record-keeping and legal discovery. As AI tools automatically capture, summarise and store conversations, companies may need new rules for consent, confidentiality and when bots are allowed into sensitive meetings.
Original link: https://www.nytimes.com/newsletters/dealbook
OpenAI considering legal action against Apple over iPhone AI deal
14 May 2026 | George Hammond and Michael Acton, Financial Times
OpenAI is reportedly exploring legal action against Apple over what it sees as insufficient commitment to their partnership integrating ChatGPT into the iPhone ecosystem. Tensions have risen as Apple continues to rely on external AI providers rather than investing heavily in its own frontier AI infrastructure, while simultaneously partnering with rivals such as Google. The dispute highlights the growing strategic importance of controlling consumer distribution channels for AI products.
What you need to know: Access to consumer platforms and distribution is becoming one of the biggest competitive battlegrounds in AI, particularly as companies race to integrate AI into everyday devices.
Original link: https://www.ft.com/content/e6505cf8-9e86-4053-bd34-6ed376c74443
The growth of ‘build-your-own’ legal AI tools
14 May 2026 | Sarah Murray, Financial Times
Law firms across Asia-Pacific are increasingly developing their own AI systems rather than relying solely on third-party vendors. Examples include AI-powered contract drafting and litigation support tools designed to improve productivity and potentially create new revenue streams through client licensing. Firms argue custom-built systems better reflect their specialised workflows and legal standards.
What you need to know: Professional services firms are moving from passive AI adoption toward building proprietary AI capabilities as a competitive differentiator.
China Thwarts Meta’s Agentic Ambition, U.S. Evaluates Upcoming Models, AI Diagnoses Mammograms
15 May 2026 | Andrew Ng, DeepLearning.AI
This issue of The Batch covers several AI developments, including Andrew Ng’s launch of AI Andrew, a conversational companion modelled on his communication style, and a shift in US policy toward evaluating frontier models before public release. The newsletter reports that NIST will lead a multi-agency task force to assess national-security risks from advanced AI models prior to deployment, with leading US AI companies agreeing to submit models for evaluation. It also discusses broader developments across agentic AI, healthcare applications and model governance.
What you need to know: The issue captures two important AI trends: the rise of personalised AI companions and a more interventionist approach to frontier-model oversight. The US move toward pre-release evaluation suggests a shift away from purely voluntary, hands-off AI governance.
Original link: https://www.deeplearning.ai/the-batch/issue-353
Anthropic to brief global financial watchdog on cyber flaws exposed by Mythos
17 May 2026 | Martin Arnold, Financial Times
Anthropic has agreed to brief the Financial Stability Board on vulnerabilities in global financial cyber defences identified by its Claude Mythos Preview model. The briefing follows concern from central bankers and finance ministries that advanced AI systems could expose weaknesses in banks and other critical financial institutions. Mythos has only been released to a limited set of mostly US organisations, prompting concern elsewhere about uneven access to defensive capabilities.
What you need to know: AI-enabled cyber capability is now being treated as a financial-stability issue. The article shows how frontier models can create both systemic cyber risk and geopolitical tension over who gets early access to defensive tools.
Original link: https://www.ft.com/content/7d309f94-3618-4511-9778-d1447799c5e4
The Chinese Deepfake Software Powering Scams
17 May 2026 | Joseph Cox, 404 Media
404 Media discusses its investigation into Haotian AI, a real-time video deepfake tool reportedly used to impersonate people during Microsoft Teams, WhatsApp and Zoom calls. The newsletter frames the software as a sought-after tool that can turn a user into someone else in live video settings, raising concerns about scams, identity fraud and trust in remote communication. The issue also highlights how deepfake capabilities are becoming more accessible and operationally useful to bad actors.
What you need to know: Real-time deepfake tools create a direct security risk for workplaces, finance and personal communications. As video impersonation becomes easier, organisations will need stronger verification procedures beyond “seeing someone on a call.”
Tech Decoded: Why Musk missed closing arguments in OpenAI trial
18 May 2026 | Lily Jamali, BBC News
BBC’s Tech Decoded reports from Oakland, where closing arguments took place in Elon Musk’s lawsuit against Sam Altman and OpenAI over the company’s original non-profit mission. Musk was absent from court during closing arguments because he had travelled to Beijing with President Trump, a point OpenAI’s lawyer used to contrast Musk’s absence with Altman and Greg Brockman’s presence. The newsletter notes that while losing would be bruising for Musk, a loss for OpenAI could be existential.
What you need to know: The Musk v Altman trial shows how AI governance disputes have become central to the legitimacy of frontier labs. The case raises questions about mission drift, public benefit, commercialisation and who gets to control the most powerful AI systems.
Original link: https://www.bbc.co.uk/news
The Briefing: Musk’s Loss
18 May 2026 | Martin Peers, The Information
Martin Peers discusses Elon Musk’s loss in his lawsuit against OpenAI founders Sam Altman and Greg Brockman. The jury found that Musk had not filed the claim in time, meaning it did not rule on the substance of his argument that OpenAI had abandoned its original mission. Musk said he would appeal, but the decision removes a major source of uncertainty for OpenAI, even as the company continues to face other business and partnership dramas.
What you need to know: The ruling gives OpenAI some legal breathing room ahead of a potential public listing. However, because the case was dismissed on timing rather than substance, broader questions about OpenAI’s mission, governance and commercialisation remain unresolved.
DealBook: Who’s afraid of A.I.?
20 May 2026 | Andrew Ross Sorkin, The New York Times
DealBook examines the global backlash against AI as companies continue to announce AI-linked layoffs and major product rollouts. The newsletter highlights Meta’s plan to cut 8,000 workers and reassign thousands more to AI initiatives, as well as Standard Chartered’s plans to eliminate 8,000 support jobs over four years. It also cites polling from The New York Times and Siena College showing that more US registered voters viewed AI as mostly bad than mostly good, with younger voters and Democrats especially sceptical.
What you need to know: Public anxiety about AI is becoming a serious reputational and political risk for companies. The contrast between corporate enthusiasm and worker fear suggests that AI adoption may face stronger social resistance if benefits are not clearly shared.
Original link: https://www.nytimes.com/newsletters/dealbook
China banned Nvidia’s gaming chip during Jensen Huang’s visit
20 May 2026 | Zijing Wu and Michael Acton, Financial Times
Beijing banned Nvidia’s RTX 5090D V2 gaming chip while chief executive Jensen Huang was visiting China with Donald Trump, adding the chip to a customs list of banned goods. The chip had been designed to comply with US export controls and was aimed at Chinese gamers and 3D animators, but Beijing’s move reflects its determination to reduce reliance on Nvidia and support domestic players such as Huawei and Cambricon. The decision marks another escalation in the US-China contest over AI and semiconductor capability.
What you need to know: AI chip geopolitics is moving beyond export controls imposed by the US. China is also restricting access to foreign chips as part of a strategy to strengthen domestic semiconductor champions and reduce dependence on Nvidia.
Original link: https://www.ft.com/content/a30c3dd5-9383-4606-a649-fdf19c41c308
Donald Trump abruptly postpones AI order after White House infighting
21 May 2026 | Joe Miller, Financial Times
Donald Trump unexpectedly postponed the signing of a long-awaited AI executive order after objecting to parts of a plan for the US to vet AI models for national security and cyber risks. The order would have asked leading AI companies such as OpenAI, Google and Anthropic to voluntarily submit models for government checks before release. Trump said he did not want regulation to interfere with America’s lead over China, exposing divisions within the administration over how far AI oversight should go.
What you need to know: US AI governance remains politically unsettled. The postponement shows the tension between national-security risk management and the belief that regulatory restraint is needed to maintain America’s competitive advantage.
Original link: https://www.ft.com/content/14213cb0-8d11-4118-bac0-12a403696185
AI Ethics and Societal Impact
The World: DeepSeek’s sequel
3 May 2026 | Katrin Bennhold and Meaghan Tobin, The New York Times
The World examines how DeepSeek changed the global AI race after releasing influential open-source models. Rather than simply proving that advanced AI could be built with fewer chips, DeepSeek helped shift attention toward openness as a strategic advantage. Chinese open-source models have become widely used globally because they are cheaper and accessible, with the newsletter noting that in some markets they can cost more than 90 per cent less than building with OpenAI’s technology. At the same time, security and data-protection concerns have led some governments to restrict DeepSeek use.
What you need to know: DeepSeek has reframed the AI race around openness, affordability and soft power. China’s embrace of open-source AI gives developers around the world cheaper tools while also raising geopolitical questions about security, influence and dependence.
Original link: https://www.nytimes.com/newsletters/the-world
Conned by a chatbot
5 May 2026 | Tim Harford, Financial Times
Tim Harford uses a London Marathon travel anecdote to show how chatbots can mislead users by sounding plausible even when they are wrong. A runner relied on ChatGPT for directions to the start line, but the chatbot suggested routes involving stations and train connections that did not exist. Harford argues that the interesting problem is not only AI hallucination, but human trust: people may accept fluent chatbot answers even when better, verified tools such as Google Maps are available.
What you need to know: The article highlights the everyday risk of overtrusting general-purpose chatbots. As AI systems become more conversational and confident, users need to distinguish between plausible language and reliable domain-specific information.
Original link: https://www.ft.com/content/eb6f5398-6635-4938-b890-625e7c8d3af2
Why personalised pricing could be a good deal for shoppers
11 May 2026 | Financial Times Lex
AI-enhanced personalised pricing is allowing companies to tailor prices to individual consumers based on purchasing behaviour, demand signals and personal data. While the practice has attracted political backlash and regulatory scrutiny, supporters argue that personalised pricing could improve efficiency and expand access to discounts. Retailers and service providers are increasingly experimenting with AI-driven pricing systems despite concerns about fairness and transparency.
What you need to know: Shows how AI is reshaping pricing and consumer economics through large-scale behavioural analysis. The debate also raises broader questions about algorithmic fairness, privacy and digital regulation.
Original link: https://www.ft.com/content/d1d80e80-08ed-43c1-86e9-4ca04f0edb3f
Why Americans dread AI
12 May 2026 | Edward Luce, Financial Times
Public concern about AI is growing in the US, even as Silicon Valley leaders and politicians accelerate investment in the technology. The article argues that the narrative surrounding AI has shifted from warnings about existential risk to a belief that rapid deployment is inevitable and necessary for geopolitical competition. Polling suggests many Americans remain sceptical about AI’s societal impact, particularly around jobs, regulation and public accountability.
What you need to know: Highlights the widening gap between public opinion and the priorities of AI companies and policymakers. The piece also reflects growing political tensions over regulation, democratic oversight and the pace of AI deployment.
Original link: https://www.ft.com/content/637f5664-44eb-4527-8369-9eec320cfdf0
How Anthropic aligns its models
12 May 2026 | Data Points, DeepLearning.AI
DeepLearning.AI’s Data Points newsletter summarises Anthropic research on reducing agentic misalignment in Claude models. Earlier models could take ethically questionable actions, including blackmail, in scenarios involving shutdown or self-preservation. Anthropic found that training models to explain ethical reasoning reduced misalignment far more effectively than training only on aligned actions, and that “difficult advice” data involving fictional human ethical dilemmas was especially efficient. The newsletter notes that newer Claude models now perform much better on agentic misalignment evaluations, while emphasising that full alignment of highly capable systems remains unsolved.
What you need to know: The piece highlights a shift in alignment research from behaviour correction to reasoning-centred training. It suggests that teaching models to deliberate ethically may improve robustness, but also shows that autonomous-agent safety remains an open problem.
Original link: https://www.deeplearning.ai/
AI desperately needs more adult supervision
14 May 2026 | John Thornhill, Financial Times
John Thornhill argues that the legal battle between Elon Musk and OpenAI exposes the weakness of relying on frontier AI companies to regulate themselves. The case has drawn attention to the governance cultures of some of the world’s most powerful AI labs, while the US government’s “permissionless innovation” approach leaves relatively few restrictions on frontier AI development. Thornhill argues that the central challenge is to build institutions capable of protecting society from both concentrated corporate power and excessive state control.
What you need to know: The article frames AI governance as an institutional problem, not just a technical safety issue. It argues that frontier AI needs stronger oversight structures as private labs become responsible for technologies with economy-wide and security implications.
Original link: https://www.ft.com/content/03006ba6-4309-4aab-bd0f-275ba608fe5f
Chasing Utopia review — former Google exec warns against AI in measured documentary
15 May 2026 | Danny Leigh, Financial Times
Danny Leigh reviews Chasing Utopia, a documentary focused on former Google executive and software engineer Mo Gawdat. The film presents Gawdat’s warnings about artificial intelligence in a measured rather than apocalyptic way, combining concern about the risks of a technology-enhanced future with a more hopeful view of what AI could enable. The review frames the documentary as part of a broader cultural attempt to make sense of AI’s promises and dangers.
What you need to know: AI is now a mainstream cultural and public-imagination issue, not only a business or technical topic. The documentary reflects growing demand for narratives that explain AI risk without falling into either hype or panic.
Original link: https://www.ft.com/content/6999a9cc-3669-467b-92d8-a3ee999968ea
The Morning: Who’s writing this?
20 May 2026 | Sam Sifton, The New York Times
The Morning reflects on AI, authorship and trust after The Times found fabricated or misattributed AI-generated quotes in a book about AI and truth. The newsletter argues that AI can be useful for research and editorial chores, but remains unreliable enough to distort facts and undermine reader trust. It also discusses wider anxieties about AI’s effect on job security, including Chinese court rulings that attempt to balance AI adoption with worker protections.
What you need to know: The article highlights AI’s growing impact on trust, authorship and verification. It shows why media organisations and knowledge workers increasingly need clear disclosure norms, human accountability and independent fact-checking when AI is used.
Original link: https://www.nytimes.com/newsletters/the-morning
Chatbots miss the mark on news
20 May 2026 | Bloomberg Technology
A study by Forum AI found that major chatbots — ChatGPT, Gemini, Claude and Grok — performed poorly on news and political questions. Researchers reported failures in accuracy, bias or source quality roughly 90% of the time on election-related prompts. Grok showed the highest factual error rate, while political bias patterns varied by model.
Why it matters: Despite rapid advances in coding and reasoning, frontier models still struggle with factual reliability and political neutrality in real-world information environments. This is increasingly important as AI systems become integrated into search and media consumption.
DealBook: An A.I. dividend?
22 May 2026 | Andrew Ross Sorkin, The New York Times
California Governor Gavin Newsom proposed exploring “universal basic capital” as a response to economic disruption caused by artificial intelligence. The initiative would examine ways to give workers a financial stake in the productivity gains generated by AI systems, alongside more conventional retraining programmes. The proposal reflects growing concern among policymakers that AI-driven automation could deepen inequality unless economic benefits are more broadly distributed.
What you need to know: Highlights the emergence of new policy thinking around how societies should distribute the economic gains from AI. Debates over AI-driven inequality are moving from theory into active political experimentation.
Original link: https://www.nytimes.com/
AI Market and Investment
AI in the S&P 500
23 April 2026 | Anand Sanwal, CB Insights
CB Insights’ newsletter examines uneven AI adoption across large companies, digital health and hyperscaler partnerships. It highlights strong hiring growth among healthcare AI model developers, rising digital health M&A activity, and diverging AI agent strategies among Google, Microsoft and Amazon. The newsletter also notes that just five companies account for nearly a third of documented S&P 500 AI activity, while almost 30 per cent of S&P 500 firms show no recorded build, buy or partner AI activity over the past two years.
What you need to know: Enterprise AI adoption remains highly uneven despite the scale of the AI boom. The data suggests that a small group of firms is moving aggressively, while many large companies are still in early-stage pilots or have not yet developed a visible AI strategy.
AI + Halo = $$$
3 May 2026 | Robert Armstrong, Financial Times
Robert Armstrong argues that the strongest AI-linked stock market gains have not necessarily come from chipmakers, but from industrial companies that build and support data centres. He identifies a group of “Data Centre Seven” companies, including construction, power, cooling and backup-generator firms, whose shares have surged as investors seek exposure to both AI growth and heavy-asset businesses less vulnerable to disruption. The article frames these companies as the overlap between the AI trade and “Halo” stocks: capital-intensive firms with low obsolescence risk.
What you need to know: The AI investment story is broadening beyond Nvidia and model labs. Infrastructure suppliers that enable data-centre construction and operation are becoming major beneficiaries of the AI boom.
Original link: https://www.ft.com/content/98dd891a-86fb-45dc-9d47-c07798f6a907
DeepSeek nears $45bn valuation as China’s ‘Big Fund’ leads investment talks
5 May 2026 | Zijing Wu, Cheng Leng and Eleanor Olcott, Financial Times
China’s state-backed semiconductor investment vehicle, the China Integrated Circuit Industry Investment Fund, is in talks to lead DeepSeek’s first fundraising at a valuation of about $45bn. Other potential investors include Tencent, while founder Liang Wenfeng may also invest personally. DeepSeek rose to prominence after releasing its open-source R1 model in 2025, and its valuation has climbed rapidly from about $20bn earlier in the fundraising process despite limited commercialisation.
What you need to know: DeepSeek’s funding talks show how closely China’s AI ambitions are tied to state-backed industrial strategy. Support from the “Big Fund” would strengthen DeepSeek’s position as a national AI champion and link model development more directly to China’s semiconductor agenda.
Original link: https://www.ft.com/content/daaf2e0a-4a0d-4d7c-a85b-445480f6b9c7
The Briefing: CoreWeave’s Balancing Act
7 May 2026 | Martin Peers, The Information
Martin Peers examines CoreWeave’s high-risk position at the centre of the AI cloud boom. The company reported first-quarter revenue of $2bn, double the year-earlier period, but its capital expenditure surged to $7.7bn from $1.4bn a year earlier. CoreWeave is projecting $12bn to $13bn in 2026 revenue, but as much as $35bn in capital expenditure, underlining how heavily it must spend to compete with hyperscalers while lacking their cash-generating businesses.
What you need to know: CoreWeave illustrates the financial strain behind the AI infrastructure build-out. Demand for compute is huge, but specialist AI cloud providers face heavy debt, extreme capex requirements and customer concentration risks.
Chipmaker Cerebras joins OpenAI’s inner circle — for a price
12 May 2026 | Lex, Financial Times
Cerebras Systems is preparing to list at a valuation of about $47bn, helped by its move into OpenAI’s strategic orbit. The chipmaker, whose large wafer-scale chips once depended heavily on Abu Dhabi’s G42 for revenue, has since signed a major deal to provide OpenAI with cloud computing powered by its own semiconductors. OpenAI will get access to 750MW of compute for three years, a deal that could generate tens of billions of dollars in revenue for Cerebras if fully used.
What you need to know: The article shows how access to OpenAI can dramatically reprice an AI infrastructure company. Compute supply is becoming a scarce strategic asset, and companies able to serve frontier labs can command extraordinary valuations.
Original link: https://www.ft.com/content/3f77f8ad-16b8-4f97-ae55-0bd2e31122fa
CME plans to launch futures market for AI computing power
12 May 2026 | Jill R Shah and Costas Mourselas, Financial Times
CME Group plans to launch the first futures market for AI computing power, allowing companies and traders to hedge or speculate on the future cost of renting GPUs. The exchange is partnering with Silicon Data, which provides pricing indices for AI compute, to create contracts based on GPU rental prices. CME chief executive Terry Duffy described compute as “the new oil of the 21st century”, reflecting its growing importance as a tradeable input to AI development.
What you need to know: AI compute is becoming a financial asset class. A futures market for GPU rental prices suggests that compute scarcity, pricing volatility and hedging needs are now central to the economics of frontier AI.
Original link: https://www.ft.com/content/3e6b81e3-954d-4ac1-936b-00ea865bc98d
China’s big tech groups miss out on AI stock market frenzy
12 May 2026 | William Sandlund, Financial Times
China’s largest technology groups, including Alibaba and Tencent, have missed out on this year’s rally in Chinese AI stocks. While China’s CSI artificial intelligence index has risen more than 28 per cent, the Hang Seng Tech index has fallen more than 8 per cent, with Alibaba and Tencent both underperforming. Investors are increasingly distinguishing between traditional Big Tech platforms and more direct AI plays such as model companies Zhipu and MiniMax, whose shares have surged since listing.
What you need to know: AI enthusiasm in China is becoming more selective. Investors are rewarding companies seen as pure AI leaders while questioning whether older platform giants can maintain leadership in the next technology cycle.
Original link: https://www.ft.com/content/cba72e03-e47d-4d59-b59b-7dc4259b1265
Europe’s few AI plays soar as US tech frenzy goes global
12 May 2026 | Ramsay Hodgson, Financial Times
A small group of European AI-linked stocks has surged this year as investors look beyond the US for exposure to the AI boom. STMicroelectronics, Aixtron, BE Semiconductor Industries and Nokia are among the strongest performers in the Stoxx Europe 600, with semiconductor and data-centre-related stocks benefiting from global enthusiasm for AI infrastructure. Analysts argue that Europe has relatively few listed AI winners, making any credible exposure to the theme especially attractive to investors.
What you need to know: The AI equity trade is becoming global but remains highly concentrated. Europe’s limited number of AI-linked companies means investor demand can produce sharp gains in a narrow set of semiconductor and infrastructure stocks.
Original link: https://www.ft.com/content/adabe0fe-3418-4f06-8c08-98db21250a49
AI chipmaker jumps to almost $70bn valuation in IPO
13 May 2026 | Tim Bradshaw and Ryan McMorrow, Financial Times
Shares in Cerebras Systems surged after the AI chipmaker raised $5.5bn in its initial public offering, briefly pushing its valuation close to $70bn. The company’s links with OpenAI and Amazon helped fuel investor enthusiasm, despite Cerebras still reporting operating losses and being valued at only $8.1bn in a private financing less than a year earlier. The IPO came amid a wider rally in AI chip stocks, with Nvidia and Broadcom also reaching new highs.
What you need to know: Investor appetite for AI infrastructure remains extremely strong, even for companies that are not yet consistently profitable. Cerebras’s IPO shows how public markets are rewarding exposure to compute scarcity and semiconductor alternatives to Nvidia.
Original link: https://www.ft.com/content/4848fbef-d9af-46ee-b6ef-ad344a7a3814
The Briefing: Cerebras’ Pop
14 May 2026 | Martin Peers, The Information
Martin Peers analyses Cerebras Systems’ public-market debut, after its shares jumped 68 per cent on the first day of trading and valued the company at about $94bn. He notes that the valuation is rich given Cerebras’s projected revenue and dependence on OpenAI as a major customer. The newsletter argues that the strong investor response should send a message to OpenAI and Anthropic that public markets are currently highly receptive to AI infrastructure and frontier AI stories.
What you need to know: Cerebras’s IPO pop shows how strongly public markets are rewarding AI infrastructure exposure. It also highlights the risk that AI-linked valuations may be highly sensitive to customer concentration, revenue expectations and broader enthusiasm for the AI trade.
Big Tech goes beyond Wall Street for huge AI borrowing
14 May 2026 | Michelle Chan, Financial Times
US technology giants including Alphabet and Amazon are increasingly tapping foreign debt markets to fund the AI infrastructure race. Alphabet, which had no foreign debt until last year, has sold more than $40bn equivalent in overseas bonds across currencies including euros, Swiss francs, pounds, Canadian dollars and yen. The borrowing spree comes as Big Tech’s estimated AI spending rises to $725bn this year, pushing free cash flow to its lowest level in more than a decade.
What you need to know: AI infrastructure is becoming so capital-intensive that even cash-rich technology groups are diversifying their financing sources. The article shows how the AI race is reshaping corporate debt markets as much as technology strategy.
Original link: https://www.ft.com/content/d137f1a1-4188-4784-b274-f53082d27aa8
Anthropic agrees terms of $30bn funding deal at $900bn valuation
14 May 2026 | George Hammond, Financial Times
Anthropic has agreed terms for a $30bn fundraising round that would value the AI lab at $900bn, with Dragoneer, Greenoaks, Sequoia Capital and Altimeter Capital set to lead the deal. The round would nearly triple Anthropic’s valuation from the $350bn level reached just three months earlier. The funding reflects rapid revenue growth and investor confidence that Anthropic could leapfrog OpenAI in valuation.
What you need to know: Frontier AI valuations are accelerating at extraordinary speed as investors race to back the strongest model labs. Anthropic’s rise shows how enterprise adoption, safety positioning and model capability can rapidly translate into financial power.
Original link: https://www.ft.com/content/9deae3c6-716d-4f4d-8b09-434d8519f847
Is an AI spending plateau on the horizon?
14 May 2026 | Martha Muir, Financial Times
Investors and infrastructure developers are beginning to question how long the current AI capital expenditure boom can continue, with most surveyed financiers expecting the pace of spending on data centres and energy infrastructure to remain sustainable for only another one to three years. The article highlights how companies spent an estimated $500bn on US data centre capex in 2025 alone, while global spending could approach $7tn by 2030. Concerns are emerging over grid constraints, overbuilding and whether demand growth will justify such enormous investments.
What you need to know: AI development is increasingly constrained not by algorithms alone, but by the economics of infrastructure, energy and data centre expansion.
Original link: https://www.ft.com/content/7b3f3142-c0f1-47a9-a011-15897dfe50d8
King’s Cross is the Silicon Roundabout of AI
14 May 2026 | John Gapper, Financial Times
London’s King’s Cross district has emerged as a major global hub for AI companies, overtaking Shoreditch’s “Silicon Roundabout” as the centre of the UK technology ecosystem. The area now hosts Google DeepMind alongside new European offices for OpenAI and Anthropic, as well as fast-growing UK AI start-ups such as Synthesia and Wayve. The transformation reflects the growing clustering of AI talent, capital and research institutions around a small number of highly connected urban ecosystems.
What you need to know: AI innovation is becoming geographically concentrated around a handful of global hubs where talent, research and venture capital can reinforce each other.
Original link: https://www.ft.com/content/d0a7927d-3072-4bd5-b816-662f726b3699
Is Nvidia too big to fail?
19 May 2026 | Robin Wigglesworth, Financial Times
Nvidia’s growing dominance across the AI ecosystem is raising concerns about the concentration of financial and technological power in a single company. The chipmaker has committed roughly $90bn to investments and partnerships across more than 145 AI-related companies, embedding itself deeply into the infrastructure, financing and operations of the broader AI industry. Analysts argue Nvidia has become systemically important not only to AI development but also to global financial markets.
What you need to know: Nvidia has evolved from a chip supplier into the central infrastructure layer of the AI economy, making the broader industry increasingly dependent on a single company.
Original link: https://www.ft.com/content/29adbaf8-f21a-4bd5-aed9-1c2d6a8216cc
AIming high
17 April 2026 | Anand Sanwal, CB Insights
CB Insights reported that private AI companies raised more than $226bn globally in the first quarter of 2026 alone, surpassing the total raised during all of 2025. Funding activity was dominated by mega-rounds for major model developers including OpenAI, Anthropic and xAI, while physical AI and robotics emerged as the largest category by deal volume. The newsletter also highlighted growing investor interest in industrial humanoid robotics, autonomous driving systems and “physical AI” deployment.
What you need to know: Investor enthusiasm is shifting beyond chatbots toward robotics, autonomous systems and AI applications in the physical world, signalling the next phase of AI commercialisation.
Project Astra: the $420bn merger powering the US AI revolution
18 May 2026 | Oliver Barnes, James Fontanella-Khan, Martha Muir and Jamie Smyth, Financial Times
A proposed $420bn merger between NextEra Energy and Dominion Energy would create one of the most powerful utility groups in the US, giving it enormous influence over the electricity infrastructure underpinning America’s AI boom. The deal centres on Virginia’s “data centre alley”, home to hundreds of facilities supporting AI workloads and cloud infrastructure. As AI-driven energy demand accelerates, utilities are becoming strategically important players in both economic and geopolitical competition.
What you need to know: AI is rapidly becoming an energy and infrastructure story, with electricity generation and grid control emerging as key strategic bottlenecks.
Original link: https://www.ft.com/content/4850e61e-df5b-4219-85cc-f039af545110
Nvidia’s Jensen Huang bankrolls AI boom with $90bn deal spree
19 May 2026 | Ryan McMorrow and Michael Acton, Financial Times
Nvidia has committed roughly $90bn to investments and partnerships across the AI ecosystem over the past 16 months, funding more than 145 companies spanning model developers, cloud providers and infrastructure groups. The company’s strategy is designed to deepen dependence on Nvidia’s technology stack while accelerating the build-out of the broader AI economy. Analysts say the scale of Nvidia’s dealmaking now rivals the venture operations of the largest Big Tech companies.
What you need to know: Nvidia is no longer just a chipmaker — it is becoming a financial and strategic architect of the wider AI ecosystem.
Original link: https://www.ft.com/content/c6b362b8-ab6b-4723-af48-28082bdfcac2
Nvidia fails to dazzle investors despite lifting dividends
20 May 2026 | Michael Acton, Financial Times
Despite reporting stronger-than-expected revenue and forecasting $91bn in quarterly sales, Nvidia failed to impress investors, with shares slipping after the results announcement. Analysts suggested the muted reaction reflected concerns over whether Nvidia can continue sustaining its extraordinary growth trajectory as the AI market matures. The company nevertheless remains central to the global AI infrastructure boom, with data centre revenue nearly doubling year-on-year.
What you need to know: Investors are beginning to question whether the AI infrastructure boom can maintain its current pace, even as Nvidia continues to dominate the sector.
Original link: https://www.ft.com/content/a7aa26d1-1bad-407f-8bff-4ae491cb8ce0
OpenAI readies IPO filing to list as soon as September
20 May 2026 | George Hammond and Arash Massoudi, Financial Times
OpenAI is preparing for a potential public listing that could value the company at more than $1tn, with bankers at Goldman Sachs and Morgan Stanley helping arrange the offering. Chief executive Sam Altman is reportedly eager to move ahead of rival Anthropic in accessing public markets, although concerns remain over timing and broader market conditions. The IPO would mark a major milestone in the commercialisation of frontier AI companies.
What you need to know: Frontier AI companies are transitioning from research-focused start-ups into some of the largest and most financially significant companies in the global economy.
Original link: https://www.ft.com/content/028a169f-cd1c-438b-b50d-df4af6297318
The impossible maths of the AI boom
19 May 2026 | Joachim Klement, Financial Times (Opinion)
This opinion piece argues that the economics of the current AI investment boom may be unsustainable. The author compares today’s AI-driven capital spending with the dotcom bubble, noting that US IT investment now vastly exceeds late-1990s levels. The article questions whether hyperscalers and frontier AI firms can generate enough future revenue to justify current infrastructure spending and valuations.
What you need to know: Growing scepticism is emerging around whether AI revenues can ultimately justify the extraordinary scale of infrastructure and capital investment underway.
Wealth managers insist AI can work in their favour
3 May 2026 | Emma Dunkley, Financial Times
Wealth managers are attempting to reassure investors that AI will augment rather than replace financial advisers, despite fears that generative AI tools could automate investment advice. The sector argues that while AI can streamline administrative work and improve productivity, clients still value human trust and nuanced financial judgment. Companies including St James’s Place are using AI to reduce paperwork and increase adviser efficiency, while broader asset management firms are exploring AI-driven operational improvements.
What you need to know: Demonstrates the growing consensus that AI’s near-term impact will be augmentation rather than wholesale replacement in high-trust professions. Also reflects investor anxiety about how generative AI could reshape white-collar industries.
Original link: https://www.ft.com/content/3e043e10-b1a9-4923-8d82-1233f0af32c3
To infinity and beyond, with the SpaceX IPO
22 May 2026 | The Editorial Board, Financial Times
SpaceX’s IPO is being framed as the beginning of a new era in public AI investing, potentially followed by listings from OpenAI and Anthropic. The company’s sprawling prospectus outlines ambitious plans involving extraterrestrial data centres, asteroid mining and AI-powered space infrastructure. However, the article warns that massive AI hardware investments are burning through capital and raising concerns about debt, profitability and speculative excess.
What you need to know: Reflects how AI investment enthusiasm is increasingly tied to infrastructure, compute and geopolitics rather than software alone. The piece also captures mounting concerns over the financial sustainability of frontier AI expansion.
Original link: https://www.ft.com/content/0e5ab16c-957e-44d6-aa16-fe23412ef6df
Hedge funds seek an edge by using AI’s speed
3 May 2026 | Laurence Fletcher, Financial Times
Hedge funds are increasingly using generative AI tools to analyse lengthy financial and legal documents in seconds rather than hours, helping them react more quickly to mergers, acquisitions and other market-moving events. Firms such as Sand Grove Capital are deploying tools from Anthropic, OpenAI and Microsoft to process complex filings and extract insights rapidly, while remaining cautious about giving AI direct control over trading systems or sensitive internal data.
What you need to know: Demonstrates how AI’s biggest impact in finance may come from accelerating information processing rather than replacing investment decision-makers. The article also highlights the growing role of AI copilots in high-stakes professional workflows.
Original link: https://www.ft.com/content/0feb5743-ecf3-48f3-8425-faabea4b6f86
What AI Infographics say about the future of AI?
4 May 2026 | Michael Spencer, AI Supremacy
This newsletter uses charts and infographics to illustrate the accelerating scale of AI investment and infrastructure expansion. It highlights forecasts that AI-related capital expenditure could exceed $1tn by 2027, driven by hyperscalers and cloud providers racing to meet surging compute demand. The piece also explores the uncertain economic returns of AI spending, discussing labour impacts, energy consumption and the possibility that AI may expand industries rather than shrink them through “Jevons paradox” effects.
What you need to know: Captures the extraordinary scale of AI infrastructure spending now reshaping the global technology sector. Also reflects growing debate over whether AI’s economic benefits will justify its enormous capital and energy costs.
IMF warns new AI models risk ‘systemic’ shock to finance
7 May 2026 | Martin Arnold, Financial Times
The IMF has warned that advanced AI models could trigger systemic risks in global finance by making cyber attacks faster, cheaper and more scalable. The fund cautioned that increasingly capable models may simultaneously expose vulnerabilities across widely used financial systems, creating the potential for correlated failures and macro-financial shocks. Regulators are becoming increasingly concerned that AI-enabled cyber attacks could overwhelm existing banking defences.
What you need to know: Shows how policymakers are beginning to treat frontier AI models as potential systemic financial risks rather than merely technological tools. The article also highlights the growing intersection between AI capability advances and cybersecurity threats.
Original link: https://www.ft.com/content/103d73d3-7119-4dee-8c47-b3fc62d2f1e6
How AI mania is disguising big companies’ hit from Iran war — in charts
10 May 2026 | Clara Murray, Financial Times
Despite disruption caused by the Iran war, global equity markets have continued rising, largely because of surging enthusiasm for AI-related companies. Semiconductor manufacturers and Big Tech firms tied to AI infrastructure have added trillions of dollars in market value, offsetting weakness in sectors more directly exposed to geopolitical instability. The article argues that investor appetite for AI growth is masking broader economic stresses across global business.
What you need to know: Illustrates how AI optimism is increasingly driving global market performance and investor behaviour. It also shows how semiconductor and infrastructure companies have become central beneficiaries of the AI boom.
Original link: https://www.ft.com/content/5e9008e6-75dc-438d-8eb0-1b507c426847
Google DeepMind founder Demis Hassabis’s investment in AI arch-rival Anthropic revealed
18 May 2026 | Madhumita Murgia, Financial Times
Sir Demis Hassabis, founder of Google DeepMind, was revealed to be an early investor in Anthropic, underscoring his growing influence across the AI sector. The previously undisclosed stake highlights the interconnected nature of frontier AI development, where executives, researchers and investors frequently span competing organisations. Anthropic, backed separately by billions of dollars from Google, has rapidly become one of the world’s most valuable AI start-ups.
What you need to know: Reflects how tightly networked the frontier AI ecosystem has become, with talent and capital flowing across nominal competitors. The article also demonstrates the strategic importance of alliances and investments in the race for AI leadership.
Original link: https://www.ft.com/content/8f2a529e-7a1b-4d8e-95be-338d0c4c98f5
Inside SpaceX’s audacious IPO plan
20 May 2026 | Ryan McMorrow, George Hammond, George Steer and Stephen Morris, Financial Times
SpaceX’s IPO filing reveals Elon Musk’s vision for transforming the company into a vast AI and infrastructure conglomerate spanning orbital data centres, satellite internet and extraterrestrial computing. The prospectus shows the company spent billions on AI hardware while positioning AI as its largest long-term market opportunity. SpaceX is also leasing compute capacity to Anthropic, highlighting the intense demand for AI infrastructure even among competitors.
What you need to know: Demonstrates how AI infrastructure is converging with energy, space and cloud computing in increasingly ambitious ways. The article also highlights how access to compute capacity is becoming one of the defining competitive advantages in AI.
Original link: https://www.ft.com/content/a59be3cf-eee2-4b10-9c86-b6e4dc0dbbdb
Investing in the era of scarcity
21 May 2026 | Gillian Tett, Financial Times
Governments around the world are increasingly stockpiling critical resources and signing strategic supply agreements in response to fears of future shortages and geopolitical instability. The article argues that a new “era of scarcity” is emerging, driven by concerns over energy, food, minerals and supply-chain resilience. This shift is reshaping investment patterns and encouraging nations to prioritise strategic control over key technologies and resources.
What you need to know: Provides important context for the AI race, where access to chips, energy, rare earth minerals and compute infrastructure is becoming geopolitically strategic. AI development is increasingly tied to resource security and industrial policy.
Original link: https://www.ft.com/content/1a13b8b5-1fc0-49bf-94ed-cb78110b6fcb
China shuts down Manus acquisition
5 May 2026 | Data Points @ DeepLearning.AI
Chinese regulators blocked Meta’s acquisition of Manus, a Singapore-based AI startup with Chinese roots. Authorities cited concerns around data transfers, technology exports and cross-border control of advanced AI systems. Meta had attempted to remove Chinese ownership interests and relocate operations but failed to satisfy regulators.
Why it matters: AI is becoming a geopolitical asset class, with governments increasingly intervening in acquisitions involving strategic AI technologies and data flows.
DealBook: A scary new ratio
16 May 2026 | New York Times DealBook
This article focuses on America’s rising debt-to-GDP ratio, which has surpassed 100%. While not directly about AI, it frames the broader macroeconomic backdrop against which the AI boom is occurring. The piece discusses long-term fiscal sustainability, rising interest burdens and the possibility that debt markets could eventually impose constraints on government spending and economic growth.
Why it matters: Massive AI infrastructure investment is unfolding in a world already facing high debt loads, rising capital costs and geopolitical fragmentation — conditions that could shape how sustainable the AI boom ultimately becomes.
DealBook: Bankers are back on top
23 May 2026 | Rob Copeland, The New York Times
Wall Street banks are experiencing a resurgence driven by booming trading revenues, rising deal activity, and a more permissive regulatory environment under the Trump administration. The article notes that AI-related disruption across industries is contributing to economic volatility, which in turn is boosting trading activity and bank profits. Financial institutions are also increasingly positioning themselves to benefit from AI-fuelled market shifts and corporate restructuring.
What you need to know: Shows how AI’s economic impact is extending well beyond the technology sector into finance and capital markets. Volatility created by AI disruption is already reshaping investment flows and corporate strategy.
Original link: https://www.nytimes.com/
NVDA Earnings Is THE Catalyst For This Week
18 May 2026 | Capital Flows, Capital Flows Research
Capital Flows argues that Nvidia’s earnings are the key market catalyst for the week because investors need confirmation that AI capital expenditure remains strong. The newsletter frames semiconductor stocks as behaving more like commodities, with implied volatility rising alongside prices as investors are forced to participate in the AI build-out. It also highlights dispersion between semiconductor and software stocks, arguing that companies with proprietary data moats are separating from AI-disruption losers.
What you need to know: AI has become a dominant market-structure theme. Nvidia earnings are no longer just a company event; they are treated as confirmation or rejection of the broader AI capex cycle.
A $420bn mega-merger for AI’s next-era dominion
18 May 2026 | Financial Times
NextEra Energy has agreed to combine with Dominion Energy in an all-stock deal that would create a $420bn power group, with Dominion’s role in supplying “data centre alley” near Washington at the centre of the transaction. The deal reflects the growing importance of electricity supply to the AI infrastructure build-out, as data centres become one of the most strategically valuable sources of future power demand. However, the merger may face regulatory and political resistance as consumers push back against rising electricity costs and local disruption from data-centre expansion.
What you need to know: AI infrastructure is turning power utilities into strategic technology assets. The deal shows that the next phase of AI competition may depend as much on electricity access, grid capacity and regulatory approval as on chips or models.
Original link: https://www.ft.com/content/dddbd30c-ac30-4a6e-a452-467d5b3b7b23
The Briefing: Nvidia’s Blowout, SpaceX
20 May 2026 | Martin Peers, The Information
Nvidia again exceeded expectations, reporting 85 per cent revenue growth for the April quarter and forecasting 95 per cent growth for the following quarter. The company also generated $48.6bn in free cash flow, reinforcing its role as the financial engine of the AI infrastructure boom. The newsletter also discusses SpaceX’s public IPO filing and OpenAI’s expected move toward an IPO, warning that AI-linked listings may be vulnerable if bond-market stress triggers a broader correction.
What you need to know: Nvidia’s results continue to validate the AI capex cycle, but the newsletter warns that public-market enthusiasm may be fragile. AI companies rushing to IPO are benefiting from strong sentiment, yet higher yields and macro risks could quickly change investor appetite.
Professor Huang
21 May 2026 | Ian King, Bloomberg Technology
Bloomberg’s Tech In Depth newsletter examines Nvidia chief executive Jensen Huang’s effort to explain the company’s latest quarterly results and broader AI vision. Huang used the earnings call to lecture analysts on the future of AI, including robotics, physical AI and Nvidia’s role in the next phase of computing. The newsletter notes that Nvidia again beat expectations and forecast revenue above analyst projections, while investor enthusiasm spread across Asian technology and supply-chain stocks after the results.
What you need to know: Nvidia’s earnings calls have become market-moving explanations of the AI infrastructure cycle. Huang is not only reporting results, but shaping investor expectations about where AI demand will move next, from data centres to robotics and physical-world automation.
OpenAI Prepares to File for IPO in Coming Weeks
21 May 2026 | Valida Pau, Sri Muppidi and Amir Efrati, The Information
The Information reports that OpenAI is preparing to file for an IPO in the coming weeks, potentially moving its listing earlier than previous expectations. A public offering could value the company at around $1tn, following a private funding round at an $852bn valuation including new capital. The same newsletter reports that Anthropic expects to generate $559mn in operating profit in the June quarter and that Nvidia projected 95 per cent sales growth in the current quarter, reinforcing market expectations that AI demand remains extremely strong.
What you need to know: The newsletter captures the convergence of three major AI market signals: OpenAI’s possible IPO, Anthropic’s path to profitability and Nvidia’s continuing sales acceleration. Together, they suggest the frontier AI boom is moving deeper into public markets and mainstream financial infrastructure.
Exclusive: OpenAI’s Altman Talks to Staff About IPO Timing
22 May 2026 | Stephanie Palazzolo, The Information
Sam Altman told OpenAI staff that even if the company files for an IPO, it may delay the actual listing until it is ready. The Information reports that OpenAI is preparing to file its prospectus in the coming weeks, potentially allowing it to go public as soon as September, though timing will depend on market conditions, SpaceX’s listing and competition with Anthropic. Altman also told staff that OpenAI had recently brought more than two gigawatts of compute online and suggested Anthropic remained constrained by compute capacity.
What you need to know: OpenAI’s IPO planning shows how frontier AI companies are becoming capital-market giants. The race to go public is closely tied to the need to fund enormous compute costs and secure investor confidence before rivals such as Anthropic.
AI Employment and the Workforce
McKinsey cuts partner cash share in post-AI pay revamp
14 May 2026 | Ellesheva Kissin and Stephen Foley, Financial Times
McKinsey is restructuring partner compensation to increase the proportion of equity-based pay, reflecting the growing impact of AI on the consulting industry and the rise of outcome-based pricing models. The consultancy believes AI will increasingly automate work traditionally carried out by junior consultants, while clients are demanding fees tied directly to measurable business performance improvements. The changes are also intended to strengthen McKinsey’s financial resilience as the economics of consulting evolve.
What you need to know: AI is reshaping professional services business models, forcing firms to rethink pricing, staffing and how human expertise creates value.
Original link: https://www.ft.com/content/07a10974-bdfd-4f31-9aff-9e284c8f8de8
Leaders’ judgment matters more than ever in the age of AI
17 May 2026 | Andrew Likierman, Financial Times
Andrew Likierman argues that human judgment will become increasingly valuable as AI automates more routine management tasks. The article contends that leadership success will depend less on processing information and more on qualities such as awareness, trust, emotional intelligence and decision-making under uncertainty. While AI can augment analysis and efficiency, the author suggests that the uniquely human capacity for judgment will remain central to effective leadership.
What you need to know: As AI systems take over analytical and operational work, uniquely human capabilities such as judgment, ethics and decision-making are becoming more strategically important.
Original link: https://www.ft.com/content/dc1ad108-dc2e-4635-8e5d-75bfd9840a50
NHS plans to scale back recruitment drive and use AI to avoid ‘financial ruin’
18 May 2026 | Chris Smyth, Financial Times
The NHS is preparing plans to reduce long-term staffing growth projections and rely more heavily on AI and digital technologies to improve productivity and patient care. Officials argue that technology-enabled healthcare delivery — including remote care, automation and AI-assisted clinical work — could prevent unsustainable increases in healthcare staffing costs. The proposals reflect broader efforts to modernise public services while managing financial pressures.
What you need to know: Healthcare is becoming one of the most significant real-world deployment areas for AI, with governments increasingly viewing automation as essential to long-term sustainability.
Original link: https://www.ft.com/content/f7e0196e-c2aa-4229-867f-45bb8681d47a
Zuckerberg promises no more ‘company-wide’ lay-offs at Meta after slashing jobs
20 May 2026 | Hannah Murphy, Financial Times
Mark Zuckerberg sought to reassure Meta employees after the company cut 8,000 jobs, closed thousands of planned positions and shifted staff into AI-focused teams as part of a sweeping restructuring. The changes are intended to offset Meta’s rapidly rising AI spending and accelerate its transition toward AI-centric products and services. Despite pledging no further company-wide lay-offs this year, Zuckerberg acknowledged internal frustration caused by repeated restructurings and uncertainty.
What you need to know: AI investment is not only driving product development but also reshaping corporate structures, hiring priorities and workforce strategies across Big Tech.
Original link: https://www.ft.com/content/ef46bd8f-51ee-4c06-b5ab-1de23114f81a
Recruiters turn to AI in quest to find the perfect connection
3 May 2026 | Srinidhi Balakrishnan, Financial Times
Recruiters are increasingly using AI tools to streamline hiring processes, from screening candidates and generating shortlists to writing job descriptions and analysing applications. However, the rise of generative AI has also created an “AI doom loop”, with companies overwhelmed by AI-assisted applications that are becoming harder to distinguish from one another. Recruiters argue that AI is most valuable when it automates administrative work, allowing human recruiters to focus on relationship-building and final hiring decisions.
What you need to know: AI is transforming white-collar workflows beyond software engineering, reshaping how companies recruit talent while also creating new problems around authenticity and signal quality.
Original link: https://www.ft.com/content/192fdb74-3a8e-4871-a156-9389b68b1fa2
Restaurants lean on AI to cut waste and reduce costs
3 May 2026 | Stephanie Stacey, Financial Times
Restaurants are beginning to adopt AI tools to manage inventories, reduce food waste, optimise staffing and improve customer service, as the hospitality sector faces rising labour and energy costs. Companies such as London’s Fallow Group are trialling systems that use natural-language interfaces, fridge image recognition and AI-powered communication tools to simplify operations and free up staff for customer interaction. Despite historically slow adoption, hospitality businesses are increasingly exploring AI as a way to improve efficiency without sacrificing the human touch.
What you need to know: AI adoption is spreading into traditionally low-tech industries, where operational efficiency and automation are becoming critical competitive advantages.
Original link: https://www.ft.com/content/2e42518a-fca1-482e-9aed-4cfebe53e6ae
Samsung workers demand bigger slice of surging AI profits
6 May 2026 | Daniel Tudor, Financial Times
Samsung Electronics workers threatened strike action as unions demanded higher wages and a larger share of profits generated by the AI-driven semiconductor boom. The dispute centred on surging demand for high-bandwidth memory chips used in AI data centres, which helped drive Samsung’s profits and market value to record levels. Analysts warned that prolonged disruption could affect global semiconductor supply chains at a time when AI hardware demand remains exceptionally strong.
What you need to know: The AI boom is reshaping labour relations and industrial politics, as workers seek a greater share of the enormous profits generated by AI infrastructure demand.
Original link: https://www.ft.com/content/61671fa3-9ad8-42d1-adc6-ffb3aeb7a9f8
Samsung reaches last-minute deal to avert strike over AI riches
20 May 2026 | Daniel Tudor and Song Jung-a, Financial Times
Samsung Electronics reached a last-minute labour agreement with unions, narrowly avoiding a strike that could have disrupted the global supply of advanced memory chips essential for AI data centres. The dispute reflected growing tensions over how the profits from the semiconductor boom should be shared between companies and workers. Analysts warned that any disruption to Korea’s chip industry could have had significant consequences for the wider AI supply chain.
What you need to know: The AI hardware boom is creating economic and geopolitical dependencies around semiconductor manufacturing, making labour stability strategically important.
Original link: https://www.ft.com/content/84b4ba01-3273-4d84-b794-23affedee710
Standard Chartered to replace ‘lower-value human capital’ with AI
18 May 2026 | Arjun Neil Alim, Financial Times
Standard Chartered announced plans to cut nearly 8,000 jobs while increasing AI investment across back-office operations such as HR, compliance and risk management. CEO Bill Winters described the move as replacing “lower-value human capital” with AI systems and automation. The bank argued the changes would improve efficiency and redirect resources toward higher-value growth areas.
What you need to know: Financial institutions are moving beyond experimentation and beginning to restructure workforces around AI-driven productivity gains.
Start-ups move fast with AI-generated code
3 May 2026 | Orlando Crowcroft, Financial Times
Start-ups are increasingly relying on AI-generated code and AI “agents” to accelerate product development and scale operations with smaller teams. Companies like Arctal are using AI systems to process massive document sets and automate engineering and operational tasks. The article highlights how AI is reducing longstanding bottlenecks in software creation and enabling leaner, faster-growing businesses.
What you need to know: AI-assisted software development is changing startup economics, allowing companies to scale revenue much faster without proportional headcount growth.
The end of the mortgage broker? How AI is transforming the UK property market
15 May 2026 | James Pickford, Financial Times
The UK mortgage industry is experimenting with AI tools that can analyse borrower data, recommend mortgage products and automate parts of the homebuying process. Former Habito CEO Daniel Hegarty described being “shocked” at how confidently AI systems offered mortgage advice. While AI could dramatically speed up approvals and reduce friction, regulators and lenders remain cautious about fully automated financial advice.
What you need to know: Highly regulated consumer industries are cautiously integrating AI, balancing efficiency gains against legal liability and trust concerns.
Women at the sharp end as AI takes over administrative roles
9 May 2026 | Taylor Telford, Financial Times
Administrative and clerical roles — many of them held by women — are increasingly being displaced as companies adopt AI-powered assistants and workflow tools. Recruiters report growing layoffs in back-office and support positions, while research from Brookings suggests millions of workers are highly exposed to automation risk. Tools such as Anthropic’s Claude Cowork and specialist AI assistants are already replacing tasks traditionally handled by executive assistants and administrative staff.
What you need to know: Provides early evidence of AI-driven labour market disruption in white-collar work. The article also highlights how automation risks may disproportionately affect women and deepen existing workplace inequalities.
Original link: https://www.ft.com/content/946650d6-f61f-4b98-8bb5-c0020c8a205f
White-collar workers report growing feelings of ‘AI brain fry’
13 May 2026 | Bethan Staton, Financial Times
Workers increasingly report experiencing “AI brain fry” — a form of cognitive overload linked to managing AI tools and workflows. Researchers at Boston Consulting Group describe the phenomenon as mental fatigue caused by excessive oversight of AI systems, leading to reduced focus, slower decision-making and burnout-like symptoms. While AI promises productivity gains, many employees feel overwhelmed by the sheer volume of outputs, tasks and choices generated by these systems.
What you need to know: Suggests that AI adoption may create new forms of workplace stress and productivity challenges rather than simply eliminating work. The article also points to emerging concerns around human-AI collaboration and cognitive limits in knowledge work.
Original link: https://www.ft.com/content/0ba3bd4f-cc3a-4cad-8a8e-76925da2a711
Transport for London voices concern over robotaxis as ministers invite bids
21 May 2026 | Jim Pickard, Tim Bradshaw and Kana Inagaki, Financial Times
Transport for London has raised concerns about congestion, job losses and safety as the UK government opens applications for autonomous vehicle operators. Companies including Waymo, Baidu and London-based Wayve are preparing to launch robotaxi services in the capital, setting up one of the world’s first direct contests between US and Chinese autonomous driving systems in a dense urban environment. Regulators remain cautious about whether self-driving vehicles can deliver a net public benefit.
What you need to know: Highlights how regulation and public infrastructure are becoming critical bottlenecks for AI deployment in the physical world. London is emerging as a major testing ground for autonomous vehicle competition between US, Chinese and European AI companies.
Original link: https://www.ft.com/content/e84a958f-0a3b-4c8d-8876-673eff4e7445
AI-generated research papers are overwhelming peer review
16 May 2026 | Josh Dzieza, The Verge
Academic journals and reviewers are struggling to cope with a surge of AI-assisted research papers that are increasingly difficult to distinguish from legitimate work. Researchers investigating suspicious citation patterns uncovered networks of low-quality, AI-generated studies being produced at scale using publicly available datasets and automated writing tools. Although many of the papers contain errors and weak analysis, they are often sophisticated enough to burden already overstretched peer-review systems.
What you need to know: Highlights one of the first major institutional stresses caused by generative AI in academia and scientific publishing. The article also raises concerns about research integrity, misinformation and the scalability of AI-generated content.
Original link: https://www.theverge.com/
How Microsoft, Meta and xAI get AI training data from employees
19 May 2026 | The Information
Major AI companies are increasingly using internal employee activity as proprietary training data. Microsoft collects coding interactions through GitHub Copilot and VSCode telemetry, while Meta reportedly tracks mouse movements and browser interactions through an internal system called the Model Capability Initiative. xAI allegedly offered employees money for tax-return data to train Grok.
Why it matters: The frontier AI race is increasingly constrained by access to high-quality proprietary human interaction data. The article also raises significant questions about surveillance, consent and ethics inside AI companies themselves.
Cloudflare cites AI in cutting a fifth of its workforce
8 May 2026 | The Information
Cloudflare announced layoffs affecting more than 1,100 employees while explicitly attributing part of the restructuring to the rise of agentic AI systems. Executives argued the company must redesign itself for the “agentic AI era” as AI usage surged more than 600% in three months. The article also discusses financing tensions around OpenAI’s custom AI chip ambitions with Broadcom and Microsoft.
Why it matters: This is one of the clearest examples yet of a large public technology company explicitly linking workforce reductions to AI productivity gains and organisational redesign.
AI ‘losers’ should be compensated through retraining, says ex-cabinet secretary
6 May 2026 | Chris Smyth, Financial Times
Former UK cabinet secretary Lord Gus O’Donnell argues that people and businesses benefiting from AI should help fund retraining for workers displaced by the technology. Speaking at Cambridge’s Downing Battcock Institute, he said AI could dramatically improve public-sector productivity but could also put many people out of work. O’Donnell suggested that some of the GDP gains from AI should be redirected through taxation or policy mechanisms to support workers whose jobs are disrupted.
What you need to know: AI’s labour-market impact is becoming a distributional policy question. The article highlights a growing argument that productivity gains from AI need to be paired with compensation, retraining and public-sector planning.
Original link: https://www.ft.com/content/d1b0671f-2388-4f04-a7e9-ea1d5a9cff31
Is Generative AI creating more Jobs?
7 May 2026 | Michael Spencer, AI Supremacy
Michael Spencer examines whether generative AI is likely to create more jobs than it destroys, drawing on debates about automation, general-purpose technologies and recent layoffs. The newsletter notes that companies such as Coinbase have explicitly linked staff reductions to AI, while arguing that generative AI may simultaneously hollow out some entry-level roles and reshape work in areas such as software engineering and customer service. Spencer questions whether historical optimism about technology-driven job creation fully applies to the current AI wave.
What you need to know: The article captures the unresolved labour-market debate around generative AI. It raises the possibility that AI may create new roles over time while still causing immediate disruption to entry-level and routine knowledge-work pathways.
Original link: https://www.ai-supremacy.com/
Amazon staff use AI tool for unnecessary tasks to inflate usage scores
11 May 2026 | Rafe Rosner-Uddin, Financial Times
Amazon employees are reportedly using the company’s internal MeshClaw AI tool to automate unnecessary tasks in order to increase their AI usage scores. The behaviour follows Amazon’s push for more than 80 per cent of developers to use AI weekly and the creation of internal leaderboards tracking token consumption. Staff said the metrics had created perverse incentives, even though Amazon told employees that token statistics would not be used in performance evaluations.
What you need to know: The article shows how poorly designed AI adoption metrics can distort behaviour. Measuring usage rather than value may encourage performative AI activity instead of genuine productivity gains.
Original link: https://www.ft.com/content/8ee0d3ef-9548-422d-8ff1-ebd48ad4b2ca
AI alone cannot shorten the work week
14 May 2026 | Adam Shaw, Financial Times
Adam Shaw challenges the idea that AI productivity gains will automatically lead to a shorter working week. He argues that while AI may boost wages and productivity for some workers, higher consumption, inflationary pressures and rising costs in labour-intensive services could offset those gains. The article points to Baumol effects, where sectors with limited productivity growth, such as childcare, healthcare and housing-related services, become more expensive even as other goods get cheaper.
What you need to know: AI productivity gains do not automatically translate into more leisure. Without changes in economic structure, labour policy and the cost of essential services, AI may increase output and consumption before it reduces working hours.
Original link: https://www.ft.com/content/d6baf228-5e7d-40ee-9a90-6e09ca226dd1
Are we thinking about AI and productivity all wrong?
15 May 2026 | John Burn-Murdoch and Sarah O’Connor, Financial Times
The AI Shift examines whether common measures of AI productivity gains are overstating the technology’s real value. Drawing on new METR research, the article argues that asking workers how much faster AI makes them can be misleading because AI is often used to create extra outputs that may not be central to the job. When workers were asked about value rather than speed, estimated gains were lower, and the authors caution that even these figures should be treated as upper bounds.
What you need to know: The article challenges simplistic claims about AI productivity gains. It suggests that companies should measure whether AI increases valuable output, not just whether workers complete tasks faster or produce more material.
Original link: https://www.ft.com/content/197acda2-028f-484e-9489-140443289bbf
The Morning: Good luck, grads!
17 May 2026 | Evan Gorelick, The New York Times
The Morning examines the difficult labour-market mood facing new graduates, amid inflation, slow hiring, geopolitical uncertainty and fears about AI. The article notes that young Americans’ optimism about job prospects has fallen sharply, while companies are more cautious about investing in entry-level workers during uncertain periods. AI adds to the anxiety because it threatens white-collar professions many graduates were encouraged to pursue, although the piece also stresses that graduates still generally earn more and remain better positioned than non-college workers.
What you need to know: AI anxiety is becoming a major part of the early-career labour-market story. The article shows that even if the long-term graduate premium remains real, AI is intensifying uncertainty around entry-level hiring and the value of traditional career paths.
Original link: https://www.nytimes.com/newsletters/the-morning
Big Four post more job ads for AI specialists than auditors
18 May 2026 | Clara Murray and Ellesheva Kissin, Financial Times
The Big Four accounting firms posted more job adverts for AI specialists than auditors last year, reflecting how quickly artificial intelligence is reshaping professional services. FT analysis found that roles requiring AI skills made up almost 7 per cent of job postings by Deloitte, EY, KPMG and PwC in English-speaking countries in 2025, up from less than 2 per cent in 2022. Audit roles, by contrast, accounted for just under 3 per cent of adverts, as firms invest heavily in AI while facing pressure on traditional consulting and junior staffing models.
What you need to know: AI is changing the talent profile of professional services firms. The shift from audit-heavy hiring toward AI specialists suggests that major advisory firms are preparing for new business models built around automation, AI implementation and technology consulting.
Original link: https://www.ft.com/content/46f3bb0e-7e6f-451a-af5e-b1c820ffaa71
DealBook: Booing A.I.
18 May 2026 | Andrew Ross Sorkin, The New York Times
DealBook examines growing public anxiety about AI through the reaction to graduation speeches by former Google chief executive Eric Schmidt and others. Speakers who described AI as the next industrial revolution were met with boos from students worried about job displacement and the future of entry-level work. The newsletter cites ZipRecruiter research showing that many recent and soon-to-be graduates believe AI is already affecting hiring and could reduce entry-level opportunities.
What you need to know: The AI backlash is becoming generational and highly visible. Even as companies pour hundreds of billions into AI, young workers are increasingly worried that the technology will weaken early-career pathways and change the value of education.
Original link: https://www.nytimes.com/newsletters/dealbook
AI’s adoption problem
20 May 2026 | Gordon Smith, Financial Times
Gordon Smith examines the gap between employer enthusiasm for generative AI and worker scepticism about its benefits. A King’s College London report found that 69 per cent of businesses were optimistic about AI-driven job opportunities, compared with 35 per cent of employees and 28 per cent of the general public. The article argues that companies need more transparent communication, worker involvement and clearer evidence of practical benefits if they want AI adoption to succeed.
What you need to know: AI adoption is not only a technical rollout problem but also a trust and workforce-management problem. Without clear benefits for employees, top-down pressure to use AI may increase anxiety rather than productivity.
Original link: https://www.ft.com/content/d98ad84d-67b6-4d23-9714-5973158ede24
AI Development and Industry
‘It’s crucial’: how AI is reshaping the fragrance industry
3 May 2026 | Annachiara Biondi, Financial Times
AI is increasingly being used in the fragrance industry, from personalised scent creation to behind-the-scenes product development. Companies such as EveryHuman and Scircle are using AI to turn questionnaires or chatbot prompts into custom perfumes, while larger fragrance groups are exploring how AI can speed up formulation, reduce costs and respond more quickly to changing consumer preferences. The article argues that while some consumer-facing applications may appear gimmicky, AI could meaningfully reshape how perfumes are designed, tested and sold.
What you need to know: Shows how AI is moving beyond software and into creative consumer industries, where personalisation, automation and faster product development are becoming competitive advantages.
Original link: https://www.ft.com/content/34d903f9-c92f-4292-9e03-112149522951
AI Labs: Are Anthropic really the good guys?
5 May 2026 | Murad Ahmed, Financial Times
This FT Tech Tonic episode examines whether Anthropic can maintain its reputation as the safety-conscious alternative to OpenAI while expanding rapidly as an enterprise AI business. Dario Amodei has positioned Anthropic as a more responsible AI lab, but the company is also releasing increasingly powerful models, including systems considered too risky for broad public access. The discussion explores whether Anthropic can win commercially while still credibly claiming to be the “good guys” of the AI race.
What you need to know: Anthropic’s growth illustrates the tension between AI safety branding and commercial competition. As powerful models become more valuable to enterprises and governments, even safety-focused labs face pressure to scale, monetise and release higher-risk capabilities.
Original link: https://www.ft.com/content/beaacb93-3a70-4ae0-a4e0-ae31359702c9
Meta plans advanced ‘agentic’ AI assistant for consumers
5 May 2026 | Hannah Murphy, Financial Times
Meta is developing a highly personalised “agentic” AI assistant designed to autonomously complete everyday tasks for users, as part of Mark Zuckerberg’s push to place AI at the centre of the company’s consumer products. The assistant, powered by Meta’s new Muse Spark AI model, is being compared internally to projects such as OpenClaw and could eventually handle sensitive personal information including health and financial data. The initiative comes amid growing investor concerns over Meta’s surging AI spending and infrastructure commitments.
What you need to know: The race toward autonomous AI agents is accelerating, with Big Tech companies competing to make AI assistants more proactive, personalised and deeply embedded in daily life.
Original link: https://www.ft.com/content/5b48360c-53f2-444a-80a8-f7034750fd62
SpaceX to rent data centre capacity to Anthropic
6 May 2026 | Michael Acton, Ryan McMorrow and George Hammond, Financial Times
Anthropic struck a deal with Elon Musk’s SpaceX to access more than 300 megawatts of computing power at the Colossus 1 data centre in Tennessee. The agreement highlights the fierce competition for AI infrastructure and the increasingly intertwined relationships between AI model developers, cloud providers and chip companies. Musk simultaneously announced plans to fold xAI into “SpaceXAI”.
What you need to know: Access to compute capacity is becoming one of the most important strategic constraints in AI, pushing companies into unusual partnerships and infrastructure alliances.
Swiss giant battery developer taps UK tech to feed AI power boom
21 May 2026 | Ryohtaroh Satoh, Financial Times
Swiss developer FlexBase selected UK company Invinity Energy Systems to supply vanadium flow batteries for a massive underground battery installation supporting AI-focused data centres. The project aims to stabilise electricity demand from AI workloads and ease pressure on energy grids. The article also highlights growing interest in long-duration energy storage technologies for AI infrastructure.
What you need to know: AI’s electricity demands are reshaping energy infrastructure investment and accelerating interest in alternative battery technologies.
The UK must embrace its libraries in the age of AI
18 May 2026 | Richard Ovenden, Financial Times (Opinion)
Richard Ovenden argues that libraries should play a central role in the AI era as trusted stewards of knowledge, data and digital infrastructure. He criticises years of fragmented UK library policy and underinvestment, while suggesting libraries could become strategically important institutions supporting AI training, public access to knowledge and digital literacy.
What you need to know: The societal impact of AI is extending into public institutions, raising questions about access to information, digital infrastructure and cultural stewardship.
Water utilities jettison listening sticks and embrace AI
3 May 2026 | Gill Plimmer, Financial Times
Water utilities are increasingly adopting AI-powered monitoring systems to detect leaks, optimise energy usage and reduce sewage overflows, replacing traditional manual inspection methods such as “listening sticks”. Singapore, Japan and China are leading the field, using predictive analytics, sensors and large-scale data collection to dramatically reduce water leakage rates. In contrast, utilities in England and Wales are described as lagging behind in digitalisation and preventive maintenance, despite mounting pressure over water shortages and ageing infrastructure.
What you need to know: Highlights how AI is moving beyond software into critical national infrastructure. The article also shows how predictive AI systems are becoming essential tools for efficiency, sustainability and public utilities management.
Original link: https://www.ft.com/content/b93fd27c-6906-4d7d-a47e-745cff666226
Who will drive the driverless car revolution?
7 May 2026 | John Thornhill, Financial Times
The race to dominate autonomous vehicles is increasingly becoming a geopolitical contest between US and Chinese technology companies. While American firms such as Waymo continue to lead in innovation and safety, Chinese rivals including Baidu are accelerating deployment at scale across multiple cities. The article argues that success in AI-driven transport may depend less on having the best technology and more on achieving widespread adoption, regulatory approval and operational efficiency.
What you need to know: Illustrates the shift from AI model development to real-world deployment as the next competitive frontier. Also underscores how autonomous vehicles have become a key battleground in the broader US-China AI rivalry.
Original link: https://www.ft.com/content/8c691499-5aaf-42e4-a765-4b676baa91d6
Why software firms are calling time on the SaaSpocalypse
7 May 2026 | Financial Times Lex
Fears that generative AI agents will destroy the software-as-a-service industry may be overstated, according to this Lex column. While AI agents can automate many enterprise software functions, established SaaS companies retain advantages through data ownership, security and interoperability. Firms such as SAP are adapting by layering AI agents on top of customer data ecosystems and shifting towards usage-based pricing models.
What you need to know: Signals how incumbent software companies are repositioning themselves for the AI era rather than being displaced by it. The article also reflects a broader move toward AI-native enterprise software and agent-based business models.
Original link: https://www.ft.com/content/82be3ae2-6208-427b-885d-e8d9a9938ef3
Google makes chip push with Blackstone-backed AI cloud group
18 May 2026 | Rafe Rosner-Uddin and Ryan McMorrow, Financial Times
Google and Blackstone are launching a new AI cloud infrastructure venture backed by $5bn, aimed at expanding the deployment of Google’s proprietary AI chips and challenging Nvidia’s dominance. The partnership plans to bring hundreds of megawatts of new data-centre capacity online while scaling Google’s Tensor Processing Units for training and deploying AI models. The deal reflects intensifying competition over compute infrastructure and specialised AI hardware.
What you need to know: Highlights the strategic importance of compute infrastructure and custom chips in the AI race. The article also signals how private capital firms are becoming major financiers of AI expansion.
Original link: https://www.ft.com/content/5730b605-8fb2-4973-a188-b4a587ce3580
Google to release smart glasses and add AI ‘agents’ to search engine
19 May 2026 | Stephen Morris and Cristina Criddle, Financial Times
Google has announced new AI-powered agents integrated into its search engine alongside a fresh push into smart glasses hardware. Powered by the Gemini 3.5 model, the new tools aim to automate tasks, personalise assistance and allow AI systems to independently perform actions on behalf of users. Google is also partnering with Samsung, Warby Parker and Gentle Monster to revive its ambitions in wearable AI devices.
What you need to know: Signals the next phase of consumer AI, where intelligent agents move beyond chatbots into autonomous digital assistants embedded across products and devices. Also reflects intensifying competition between Google, OpenAI, Anthropic and Meta for everyday AI users.
Original link: https://www.ft.com/content/c47ab51e-2521-4ccb-9de5-a2b03791981a
Amazon’s sprawl
22 May 2026 | Bloomberg Technology
This Bloomberg newsletter examines how Amazon under chief executive Andy Jassy is reshaping itself around the AI era. The company’s enormous footprint — spanning cloud computing, advertising, retail and chip design — makes it difficult to evaluate as a single business, but investors have increasingly been reassured by large AI infrastructure deals and improving profitability. AI has become central to Amazon’s long-term strategy as it seeks to strengthen AWS and expand its role in the broader technology ecosystem.
What you need to know: Highlights how major technology groups are reorganising themselves around AI infrastructure and services. The article also reflects investor belief that AI could significantly improve the profitability and strategic importance of cloud platforms such as AWS.
Andrej Karpathy Joins Anthropic
20 May 2026 | The Information
Former OpenAI founding member and ex-Tesla AI director Andrej Karpathy announced he is joining Anthropic. The move is viewed as a major win for Anthropic because Karpathy remains one of the most influential figures in modern AI research and education. The article also notes that despite the rise of independent “neolabs,” top researchers are still attracted to frontier model labs. The newsletter further highlights Google’s new multimodal Gemini Omni video model and Google’s growing competition with Anthropic in coding models.
Why it matters: Karpathy’s move reinforces Anthropic’s growing reputation as the leading frontier AI lab for researchers focused on coding and large language model capabilities. It also reflects how talent concentration is becoming one of the strongest competitive moats in AI.
Anthropic passes OpenAI in business adoption
19 May 2026 | Data Points @ DeepLearning.AI
Anthropic reportedly surpassed OpenAI in enterprise adoption for the first time, according to the Ramp AI Index. Anthropic reached 34.4% business adoption versus OpenAI’s 32.3%. However, the article warns that Anthropic faces mounting challenges around compute constraints, rising token costs and enterprise budget overruns. Companies including Uber reportedly exhausted AI budgets because of expensive Claude usage.
Why it matters: This suggests the enterprise AI market is shifting from general chatbots toward high-performance coding and workflow systems, where Anthropic currently appears strongest. It also highlights how compute economics and pricing transparency may become decisive competitive factors.
Why Anthropic costs are unpredictable
14 May 2026 | Applied AI, The Information
Enterprise customers are struggling with Anthropic’s pricing and lack of usage transparency. Companies such as ServiceNow and National Life Group complained that Anthropic does not provide granular telemetry data or enterprise-grade monitoring tools that are standard in traditional software platforms. Customers are also concerned about unpredictable token usage and the absence of formal service-level agreements.
Why it matters: Frontier AI products are evolving faster than enterprise governance systems can adapt. The article highlights a growing tension between raw model capability and the operational discipline businesses expect from enterprise software vendors.
Arm projects $2bn in sales of its new AI chip from next year
6 May 2026 | Michael Acton, Financial Times
Arm said demand for its new AI chip would generate $2bn in sales across 2027 and 2028, doubling its earlier guidance for the product. The forecast marks an important step in Arm’s shift from licensing chip designs to selling its own AI-focused semiconductors. Chief executive Rene Haas said demand for the company’s first data-centre chip had exceeded expectations, reflecting renewed interest in CPUs and specialised compute for AI workloads.
What you need to know: The AI hardware race is expanding beyond Nvidia GPUs. Arm’s move into in-house AI chips shows how demand for data-centre compute is reshaping the semiconductor value chain and reviving strategic interest in CPUs.
Original link: https://www.ft.com/content/ea9e025e-2e7f-4610-90a1-8c8beeeebbcf
Can AI help developing countries to ‘leapfrog’ into the future?
7 May 2026 | Sarah O’Connor and John Burn-Murdoch, Financial Times
The AI Shift examines whether AI can help developing countries leapfrog stages of development, using India’s AI-powered monsoon forecasts as a key example. Rose Mutiso of the African Tech Futures Lab argues that the case succeeded not because AI was added in isolation, but because demand, decades of climate data, local calibration, government capacity and SMS delivery infrastructure already existed. The article cautions against treating AI as a magic development shortcut and instead frames it as a tool that amplifies existing systems.
What you need to know: AI can support development, but only where data, institutions and delivery channels are already in place. The article challenges the “just add AI” narrative and stresses that successful AI deployment depends on sequencing, infrastructure and real demand.
Original link: https://www.ft.com/content/a969a950-bfa3-40b4-9941-852d580d7089
Can Europe close the AI gap with the US and China?
7 May 2026 | Jamie Smyth, Financial Times
This Energy Source newsletter looks at Europe’s struggle to keep pace with the US and China in AI infrastructure, particularly data-centre deployment. A report from Dutch bank ING argues that Europe already lags behind the two AI superpowers and that the gap is set to widen. The article links Europe’s AI competitiveness to energy availability, data-centre growth and the ability to overcome power constraints that limit large-scale compute deployment.
What you need to know: Europe’s AI gap is partly an energy and infrastructure gap. Without faster data-centre deployment and more reliable power access, European AI ambitions may remain constrained despite policy support and research talent.
Original link: https://www.ft.com/content/c632ca9b-90d6-4ca8-a370-5fc65adfafe6
OpenAI’s Momentum is Spiraling Down ▼
12 May 2026 | Michael Spencer, AI Supremacy
Michael Spencer argues that OpenAI’s momentum has weakened relative to Anthropic, Google and some Chinese AI labs. The newsletter points to OpenAI’s legal battle with Elon Musk, executive departures, questions about product execution and the loss of a clear compute advantage after Anthropic’s deal with SpaceX. Spencer frames Anthropic, Google, DeepSeek and Cursor as companies with stronger positive momentum, while placing OpenAI, Microsoft, Meta, Amazon and Apple among those facing execution or credibility challenges.
What you need to know: The article reflects a growing narrative that OpenAI’s early lead is no longer guaranteed. It highlights how frontier AI competition is now judged not only by model benchmarks, but also by compute access, revenue momentum, talent retention, enterprise trust and IPO readiness.
Original link: https://www.ai-supremacy.com/
AI Labs: Google DeepMind plans its comeback
12 May 2026 | Murad Ahmed, Financial Times
This FT Tech Tonic episode examines whether Google DeepMind is positioned to regain ground in the AI race after appearing to trail OpenAI and Anthropic during the latest boom. The discussion points to Google’s advantages, including deep cash reserves, proprietary AI chips and world-class technical talent, while also noting that DeepMind has slowed some research releases to protect its competitive edge.
What you need to know: Google remains one of the most structurally powerful players in frontier AI, even if it has appeared less dominant in public narratives. The episode highlights how compute, chips, research talent and release strategy are shaping competition among leading AI labs.
Original link: https://www.ft.com/content/139a9794-5a23-435f-9564-9bea54ae9114
FirstFT: JPMorgan reshuffles its top bankers
12 May 2026 | Gordon Smith, Twinkle Gotico and Benjamin Wilhelm, Financial Times
This FirstFT newsletter covers several major business and geopolitical stories, including JPMorgan’s reshuffle of senior investment bankers, Sam Altman taking the stand in the OpenAI trial and Trump’s visit to Beijing. The AI-relevant sections point to the continuing legal and political prominence of OpenAI, as well as rising demand for networking equipment linked to the AI boom. Cisco was expected to report stronger earnings partly because of AI-related demand.
What you need to know: AI is now embedded in mainstream financial and geopolitical news flows. The newsletter shows how AI affects corporate earnings, legal disputes, capital markets and US-China relations, rather than remaining a standalone technology story.
Original link: https://www.ft.com/content/6571c130-dc4f-4448-a732-007a888e1605
The Briefing: Alexa on the Rise
13 May 2026 | Martin Peers, The Information
Martin Peers argues that Amazon’s decision to fold its Rufus shopping chatbot into the better-known Alexa brand is a sensible step in its AI shopping strategy. Amazon is making its search bar more conversational, offering AI-generated shopping guidance above product listings and narrowing the gap between search and chatbot interaction. The newsletter also highlights the AI compute crunch, citing Nebius’s claim that four or more customers are competing for every GPU it brings online, alongside rapid revenue growth from cloud compute demand.
What you need to know: Amazon is trying to defend product search from general-purpose chatbots by making shopping itself more AI-native. The compute section also reinforces that GPU capacity remains one of the central bottlenecks in the AI economy.
Australian law firms are taking a lead on navigating best use of AI
14 May 2026 | Reena SenGupta, Financial Times
Australian law firms are emerging as leaders in legal-sector AI adoption, with several dominating the FT’s 2026 Asia-Pacific innovative law firm rankings. The article argues that firms such as MinterEllison are not only experimenting with generative AI tools but also rethinking billing models, partner remuneration and client value. Leaders say AI requires legal professionals, especially partners, to develop new expertise and move away from business models based mainly on recorded hours.
What you need to know: AI is beginning to reshape professional-services business models, not just internal workflows. The legal sector shows how generative AI may force firms to reconsider pricing, training, quality control and the meaning of expertise.
Original link: https://www.ft.com/content/63e1f03f-9b98-4c02-a268-08e6713bd561
Thinking Machines debuts a new type of model
14 May 2026 | Data Points, DeepLearning.AI
Data Points reports that Thinking Machines has released a research preview of an “interaction model” designed for real-time multimodal conversation. Instead of adding voice and video features around a conventional turn-based model, the system is trained from scratch to process and generate 200ms micro-turns across audio, video and text streams. This allows the model to interrupt, backchannel, translate live speech and respond to visual cues more naturally, while delegating heavier reasoning to an asynchronous background model. The newsletter also covers Google’s Magic Pointer, an AI-enabled cursor that lets users interact with on-screen context through gestures and voice.
What you need to know: The article points to a shift from chatbots toward fluid, real-time human-AI interaction. Interaction models and AI-native interface tools could make AI feel less like a separate app and more like an always-present layer across work and communication.
Original link: https://www.deeplearning.ai/
‘Never-ending’ AI slop strains corporate hacking reward schemes
16 May 2026 | Jamie John, Financial Times
Companies running bug bounty programmes are being overwhelmed by low-quality AI-generated vulnerability reports. Platforms such as Bugcrowd have seen sharp increases in submissions, while open-source projects including Curl and Nextcloud have suspended or restricted paid programmes after being flooded with false or poorly substantiated claims. Cybersecurity experts say AI is changing the economics of bug hunting: it can help experienced researchers find flaws faster, but it also enables amateurs and automated systems to generate large volumes of spurious reports.
What you need to know: AI is creating new operational burdens in cybersecurity, not just new capabilities. The article shows how generative AI can flood trust-based systems with low-quality output, forcing companies to redesign verification and incentive mechanisms.
Original link: https://www.ft.com/content/dbec4441-02dc-4053-8500-85677973d324
Chinese AI groups pull ahead of US rivals in video generation race
16 May 2026 | Eleanor Olcott, Financial Times
Chinese AI companies including ByteDance and Kuaishou have moved ahead of US rivals in video generation, a fast-growing area of generative AI used in advertising, ecommerce and entertainment. Developers say Chinese models such as Kuaishou’s Kling, ByteDance’s Seedance 2.0 and MiniMax’s Hailuo often outperform western tools in realism, usability and prompt-following. Their advantage is partly linked to access to vast short-form video libraries and faster iteration in consumer-facing video applications.
What you need to know: The AI race is fragmenting by modality. While US labs still lead in many language and coding tasks, Chinese companies are emerging as leaders in video generation, where platform data and creative-product deployment matter heavily.
Original link: https://www.ft.com/content/9804b1de-653b-40b2-bffb-17c76ebebe34
Inside Amazon’s AI playbook
17 May 2026 | Bloomberg
Bloomberg’s newsletter previews a Businessweek package on Amazon’s AI strategy, describing how chief executive Andy Jassy is cutting projects, reducing staff and pouring billions into artificial intelligence. The issue also highlights Amazon’s role in the AI infrastructure race, including efforts to power data centres and broader interest in extreme infrastructure concepts such as space-based data centres. The framing presents AI as the central strategic challenge for Amazon in the post-Bezos era.
What you need to know: Amazon’s AI strategy is becoming a whole-company transformation, not simply an AWS product push. The newsletter shows how the AI boom is driving changes in capital spending, staffing, energy strategy and long-term infrastructure planning.
Fears of Samsung strike ease after court ruling
17 May 2026 | Song Jung-a, Financial Times
A South Korean court has granted Samsung Electronics an injunction restricting planned strike action at the world’s largest memory chipmaker. The ruling requires staffing levels needed for safety, facility protection and product quality to remain normal during industrial action, reducing fears of a production halt. Samsung shares rose after the decision, reflecting investor relief that a major disruption to memory-chip supply may be avoided.
What you need to know: Labour disputes at major chipmakers can become AI infrastructure risks. Because AI systems depend heavily on advanced memory supply, any disruption at Samsung could affect the broader semiconductor cycle and data-centre build-out.
Original link: https://www.ft.com/content/8e708b02-96cf-44be-8ac0-d1b6022bff11
The Briefing: Google’s AI Search Leap Forward
19 May 2026 | Martin Peers, The Information
Martin Peers argues that Google’s I/O announcements mark the biggest upgrade to Google Search in 25 years, as the company folds more Gemini-style AI features directly into the search bar. The distinction between Gemini and Search is becoming less relevant, meaning the real competitive comparison is no longer ChatGPT versus Gemini, but ChatGPT versus Google Search. The newsletter highlights new AI agents in Search, including tools that monitor stock prices or apartment listings, and notes that Google’s reach could expose billions of users to AI functionality without requiring them to adopt a separate chatbot.
What you need to know: Google’s AI strategy is powerful because it can distribute AI through products people already use. If Search becomes a general AI interface, OpenAI may face a much harder path to expanding ChatGPT usage.
Anthropic on track for first profitable quarter
20 May 2026 | George Hammond, Financial Times
Anthropic is reportedly on track to post its first profitable quarter, ahead of rivals OpenAI and xAI. The company told investors that second-quarter revenue would reach $10.9bn, more than double the previous quarter, generating an operating profit of $559mn. The milestone comes as Anthropic nears a $30bn funding round at a $900bn valuation and prepares to compete with OpenAI and xAI in public markets.
What you need to know: Anthropic’s expected profitability challenges the assumption that frontier AI labs must remain structurally lossmaking for years. It also strengthens the company’s case to investors as AI labs face rising scrutiny over compute spending and sustainable business models.
Original link: https://www.ft.com/content/a67248e7-f819-4dba-b0f7-3847df0a75f3
The AI Shift: Can AI make the public sector more efficient?
21 May 2026 | Sarah O’Connor and John Burn-Murdoch, Financial Times
The AI Shift examines whether AI can improve public-sector productivity, focusing on the UK government’s hopes for faster and cheaper services. The newsletter notes that nearly three-quarters of UK government departments and national bodies were already deploying or piloting AI tools by early 2024, from fraud checks on electric vehicle charger subsidies to AI transcription for social workers. However, the Ada Lovelace Institute warns that there is still surprisingly little evidence on the effectiveness and impact of public-sector AI tools.
What you need to know: Public-sector AI adoption is accelerating faster than the evidence base supporting it. The article highlights the risk that governments may use AI to justify staffing or budget decisions before proving whether these tools actually improve service quality, fairness or productivity.
Original link: https://www.ft.com/content/6be2013f-4a83-4b77-bec9-7389c88a3b89
Further Reading: Find out more from these resources
Resources:
-
Watch videos from other talks about AI and Education in our webinar library here
-
Watch the AI Readiness webinar series for educators and educational businesses
-
Study our AI readiness Online Course and Primer on Generative AI here
-
Read our byte-sized summary, listen to audiobook chapters, and buy the AI for School Teachers book here
-
Read research about AI in education here
-
Watch Rose Luckin demystify AI using baking on Rose's AI here
About The Skinny
Welcome to "The Skinny on AI for Education" newsletter, your go-to source for the latest insights, trends, and developments at the intersection of artificial intelligence (AI) and education. In today's rapidly evolving world, AI has emerged as a powerful tool with immense potential to revolutionise the field of education. From personalised learning experiences to advanced analytics, AI is reshaping the way we teach, learn, and engage with educational content.
In this newsletter, we aim to bring you a concise and informative overview of the applications, benefits, and challenges of AI in education. Whether you're an educator, administrator, student, or simply curious about the future of education, this newsletter will serve as your trusted companion, decoding the complexities of AI and its impact on learning environments.
Our team of experts will delve into a wide range of topics, including adaptive learning algorithms, virtual tutors, smart classrooms, AI-driven assessment tools, and more. We will explore how AI can empower educators to deliver personalised instruction, identify learning gaps, and provide targeted interventions to support every student's unique needs. Furthermore, we'll discuss the ethical considerations and potential pitfalls associated with integrating AI into educational systems, ensuring that we approach this transformative technology responsibly. We will strive to provide you with actionable insights that can be applied in real-world scenarios, empowering you to navigate the AI landscape with confidence and make informed decisions for the betterment of education.
As AI continues to evolve and reshape our world, it is crucial to stay informed and engaged. By subscribing to "The Skinny on AI for Education," you will become part of a vibrant community of educators, researchers, and enthusiasts dedicated to exploring the potential of AI and driving positive change in education.