‘My job is going’: UK workers squeezed out by AI
ByAFP
May 26, 2026
The IMF estimated in 2024 that more than two-thirds of British workers perform tasks that AI could potentially carry out - Copyright AFP JUSTIN TALLIS
Lucie LEQUIER
When a client asked her a year ago to design a glossary to train an artificial intelligence system, translator Jessica Spengler realised she was going to train her own replacement.
“That was the day I really thought… my job is going,” said the 52-year-old, who translates into English for German educational and historical organisations.
In the UK, where services account for around 80 percent of the economy, AI has become flexible, fast and inexpensive competition for many white-collar workers, with the impacts beginning to emerge.
The IMF estimated in 2024 that more than two-thirds of British workers perform tasks that AI could potentially carry out, making the country more exposed than many other advanced economies.
“Some publishers have offered me lower rates than I was getting 10 years ago,” the Brighton-based Spengler told AFP, adding that she no longer receives requests to translate corporate press releases or user manuals, typically an “entry point” into the profession.
Instead, she is increasingly offered work proofreading machine-generated translations.
Translators “have to rewrite the whole thing, redo the translations, but they still only get paid the reduced rate,” said Holly Parsons, a Spanish-to-English translator at the beginning of her career.
“It’s hard as a translator to actually charge what the work is worth because people just don’t want to pay it,” the 24-year-old added.
She still earns most of her income working as a children’s activity leader.
– Change of direction –
According to a report from Morgan Stanley, British companies that adopted AI cut their workforces by eight percent in the year to October 2025 — more than in Germany, Japan or Australia.
Among the countries featured in the report, only the United States saw employment rise with AI.
“Film work has definitely been impacted by AI… it’s really kicked us down,” said Laura, 35, a director of photography in London, who preferred not to share her last name for professional reasons.
To escape the broader crisis hitting the film industry, she is retraining as an outdoor instructor in Dorset, southwest England, earning minimum wage.
After working on the short film “Mad Bills to Pay”, which won an award at the Sundance Film Festival, 35-year-old Rufai Ajala also changed direction and is now training to become a plumber.
“I’m not going to rely on film as my main focus… I don’t see it as a career option anymore where you can have stability,” Ajala said, adding that the aim was to find an “AI-proof” career.
– ‘Painful transition’ –
“There is going to be sort of a painful transition process because new jobs will take time to emerge,” said Bouke Klein Teeselink, an economics professor at King’s College London.
He said it would require “a massive adjustment for society,” which could mean “a big increase in unemployment.”
According to one of his studies, professions most exposed to AI, such as software developers and data analysts, reduced job postings after the launch of ChatGPT in November 2022, particularly for entry-level positions.
The growth of AI comes as Britain already faces high levels of youth unemployment, with the war in the Middle East and an increased minimum wage weighing on hiring.
One in six Britons aged 16 to 24 is out of work, the highest level since 2014, according to official data.
Teeselink said, however, that another market dynamic is at play with AI: productivity gains could lead to lower prices, which in turn could stimulate demand and increase employment.
He said the UK was “reasonably well positioned” for the AI transition thanks to its high-quality universities, which are set to play a crucial role in “upskilling young people to use AI well.”
ByAFP
May 28, 2026
A silicon wafer seen magnified through the lens of a microscope - Copyright AFP/File ANTHONY WALLACE
Katie Forster, with Joy Chiang in Taipei
Runaway profits and sky-high valuations for microchip companies have fuelled worker demands over pay packages in South Korea, raising the question: who profits from the artificial intelligence boom?
In the United States, some employees with stock options have made it rich and retired early, while in Asia chip engineers are now using their “immense” leverage over companies to get their way, analysts say.
After memory chip giant Samsung Electronics reached a deal with its biggest union over bonuses, averting a major strike, AFP looks at what the dispute might mean for the industry.
– Why are chipmakers suddenly so flush? –
Rapid advances in AI systems since text generator ChatGPT’s 2022 breakthrough have sparked a gold rush for tech companies.
Massive demand for the silicon components used in AI data centres — especially memory chips, which are in short supply — has sent revenues soaring for firms that design, produce and assemble them.
Samsung’s value topped $1 trillion this month, followed by Korean rival SK hynix and US chipmaker Micron — newcomers to a previously exclusive club of around a dozen companies, nearly all American.
“An unprecedented wave of insatiable demand” for advanced memory chips has made SK hynix and its peers “an indispensable backbone of the global AI infrastructure build-out”, William Keating, head of semiconductor research firm Ingenuity, told AFP.
– Do workers see the benefits? –
In the United States, employees often get stock options, a form of so-called “golden handcuffs” allowing workers to profit from share price gains over a set period of time.
Asia’s huge chip sector “is more dominated by labour unions”, Neil Shah, co-founder of Counterpoint Research, told AFP.
With Taiwan and South Korea home to most of the world’s chipmaking talent pool, engineers there hold “immense” leverage, Shah said.
“This skilled labour force know they are indispensable, they are contributing to these larger margins,” he added.
Under the union deal, around 60 percent of Samsung’s domestic workforce is eligible to receive a bonus of roughly $370,000 this year, based on a market estimate of operating profit.
Workers at SK hynix received bonuses more than three times larger than those paid by Samsung last year, according to Samsung’s union.
– Will the Samsung deal inspire others? –
A Samsung strike “would almost certainly have been the biggest work stoppage in the history of the global semiconductor industry”, South Korean writer and researcher Kap Seol said in an article for US magazine Jacobin.
In the chip world, “high pay and generous benefits often foster a sense of privilege and prestige” among workers, “despite their experience of chemically drenched working conditions, cutthroat competition, and long, risky working hours,” he wrote.
There have also been reports of discontent over bonuses at Taiwan’s chip production giant TSMC, where AI demand has brought record profits.
“As the company continues to grow, we are highly confident that the full-year growth percentage of our employee profit-sharing… will surpass that of last year,” TSMC said in a statement.
TSMC boss CC Wei held a meeting to explain the bonus situation to staff on Wednesday, and the atmosphere was “calm and friendly”, a company spokesperson told AFP, adding that annual bonuses were set to grow “more than 30 percent” year-on-year.
The Samsung agreement is fuelling labour demands in other sectors across South Korea — with workers in industries from biotech and autos to shipbuilding asking for a larger share of corporate profits through bonuses.
– Who else is cashing in? –
According to Shah, in general terms, company shareholders are reaping the most from booming profits, followed by senior company executives and then employees who have stock options.
Fourth are the actual chip engineers, some of whom are now demanding a greater share of the pie.
In the case of Californian AI chip titan Nvidia — now the world’s most valuable company at more than $5 trillion — many employees with stock options became millionaires very quickly, Shah said.
“Many of them actually left and became investors” or simply chose to retire “on the beach”.
“Taxing AI directly ties the solution directly to the problem,” wrote Rep. Greg Casar. “If AI use grows quickly, driving layoffs alongside it, the revenue from an AI tax would go up too.”
Rep. Greg Casar (D-Texas) speaks during a news conference on April 29, 2026 outside the US Capitol in Washington, DC.
(Photo by Tom Brenner/Getty Images)
Jake Johnson
May 28, 2026
C0MMON DREAMS
Two leading progressives in the US Congress are calling for a tax on artificial intelligence to fund programs that would help prevent an economic catastrophe for workers displaced by the rapidly advancing technology.
In separate op-eds published Wednesday and Thursday, Sen. Elizabeth Warren (D-Mass.) and Rep. Greg Casar (D-Texas) warned that AI risks turbocharging existing wealth and income inequality by driving up the fortunes of large companies and their executives, while hurling millions of workers into joblessness without an adequate safety net.
“Taxing AI is one way we make sure the winnings from AI benefit all Americans, rather than channeling them only to the wealthy few. If millions of people lose their jobs to AI, we’ll need the funds to deliver universal healthcare so those workers are not bankrupted by a visit to the doctor,” Warren wrote in TIME. “If AI transforms the future of work, we’ll need to invest in free education and apprenticeships and a new jobs guarantee so that all Americans have good-paying work. And while workers get back on their feet, we’ll need the revenue to bolster unemployment insurance to keep families afloat. The only way we can get there is by overhauling our tax code.”
Casar, chairman of the Congressional Progressive Caucus, made the case for an “AI tax-funded jobs program” in an op-ed for The American Prospect, arguing the initiative “should draw inspiration from the New Deal-era Works Progress Administration, which employed millions of Americans.”
The Texas Democrat specifically proposed a tax on AI “tokens,” units of data that are processed by artificial intelligence models.
“Taxing AI directly ties the solution directly to the problem,” Casar wrote. “If AI use grows quickly, driving layoffs alongside it, the revenue from an AI tax would go up too. Unlike traditional corporate taxes, an AI tax like the one I am proposing works even if employers fire workers before AI companies show a profit.”
The progressive lawmakers’ call for a new AI tax come amid mounting concerns, in the US and around the world, about burgeoning technology’s impact on workers whose jobs could be replaced by robots. Pope Leo XIV used his first encyclical to warn of the threat AI poses to employment, and prominent lawmakers such as US Sen. Bernie Sanders (I-Vt.) have predicted “economic devastation for working people” if tech oligarchs get their way.
“Corporations are already using AI to cut jobs,” Sanders’ office noted in a recent report. “Amazon, Walmart, UnitedHealth Group, JPMorgan Chase,and other companies are openly telling investors that AI will allow them to slash payrolls—even as they post tens of billions in profits and reward CEOs with pay packages of $25 million, $35 million, or more.”
Warren and Casar argued that nightmare scenarios envisioned by AI critics and industry leaders alike are entirely preventable—but averting them would require bold and urgent legislative action that’s a longshot with President Donald Trump and Republicans in control the federal government.
“Congress should act now, and not wait to see if the worst-case scenario arrives,” wrote Casar. “AI companies are already pouring hundreds of millions of dollars into elections to try to shape what regulations get considered. We cannot wait for these companies to become even wealthier and more powerful.”
Warren called for “taxing AI companies directly,” including by imposing levies on AI data centers, which have drawn grassroots backlash across the country. The senator also pushed for broader action, including a wealth tax, to ensure that mega-rich beneficiaries of the AI boom don’t “pay lower tax rates than the workers they fire.”
“Here’s what I see clearly: If we overhaul our tax code and tax AI, we can use that money to build a country that works for everyone,” Warren wrote. “A country where healthcare is treated as a human right, where every American is guaranteed a good job, and where education isn’t a privilege reserved for the wealthy. That’s what I believe taxing AI promises.”
The New Religion of AI Accelerationism
Photo by Igor Omilaev
Earlier this month, researchers at the University of California San Diego published a study offering “the first empirical evidence that a modern artificial intelligence system can pass the Turing test.” Famously named for Alan Turing, the English mathematician and World War II codebreaker, the test is designed to determine whether a computer can exhibit human intelligence such as to make it indistinguishable from a human. What is interesting, perhaps, is how few waves this apparent breakthrough has made within the broader public discourse.
Many, including Nvidia CEO Jensen Huang, reckon that we have already achieved artificial general intelligence (AGI), though there remains widespread disagreement on just what that means. There is likewise disagreement about whether the Turing test is the right one for determining whether we have AGI.
We are unlikely to perceive something like superintelligent AI taking over the world as a clear and sudden break. We’re blowing through long-awaited milestones without a real opportunity to process the implications. Within such a context of rapidly growing power and the confusion around it, it becomes important to question some easy assumptions.
AI development does not represent or grow out of neutral technological progress or “market forces.” It is a deeply coercive and political project driven by a collusive state-capitalist oligopoly and supported by an ideology that openly devalues human life. One increasingly visible proponent of this ideological complex is the English philosopher Nick Land, called “a living meme and an oracle” for the fascination his ideas have generated.
Feted as a patron saint among the Silicon Valley tech set, Land is known for popularizing a set of ideas associated with accelerationism. Though there have now sprouted dozens of variations, the core of Land’s accelerationist approach is the idea that super-intelligent AI is inherent to the dynamics of technological capitalism and ultimately can’t be stopped. He argues that AI represents capitalism’s awareness of itself, and he offers what is arguably the clearest and most well-known formulation of much of the doomerism of the present moment: “Nothing human makes it out of the near-future.” Some of the richest humans to have ever lived seem to have made their peace with this millenarian eagerness to help propel humanity into a future without humans.
Land’s version of accelerationism sees capitalism not only as a political-economic system, but as a process that intensifies and perfects itself completely on its own. The dynamics of the system, not the values of human beings, are the drivers of change and progress. Our societies and systems of values are, in this view of the world, obsolete and irrelevant. These conversations are increasingly high profile, having burst from the realm of internet obscurity onto the pages of, for example, The New Yorker.
We are told that nothing human will survive this transition, but that we should nonetheless hurry the unfolding process along. We are told that AI will aid the police state in spying on us and violating our rights, but that we should stake the U.S. (and indeed global) economy on it. We are assured that robots will displace millions or billions of human workers, but that we should herald and celebrate this in religious and eschatological terms.
These contradictions are at the center of the current conversation about AI, and they help explain why reactions to the merest mention of AI are becoming more charged with anger and resentment. Today, the stocks of the Mag 7 companies, a group of the largest and most powerful technology firms, make up 35 percent of the value of the S&P 500. Back in 2020, these companies pulled an annual return (65.8 percent) that was more than quadruple that of the S&P 500 (16.3). Every one of these companies is now worth more than $1 trillion.
Today’s technology sector does not represent the principles of anything like actual free-market competition; intensively subsidized by the public and deeply tied to the federal government, the major tech companies are a state-capital oligopoly that have benefited enormously from a variety of special subsidies and perks unavailable to ordinary companies and citizens. When we account for direct federal grants and subsidies, infrastructure support, and hardware manufacturing, public subsidies and allocations for AI have reached well into the hundreds of billions of dollars.
A Brookings Institute report published earlier this month analyzes some of the disturbing trends around the government’s relationship with the tech sector and AI technologies. The overwhelming majority of federal government procurement of AI systems takes place within the Pentagon. The Brookings report describes recent explosions in federal commitments to AI as “staggering,” showing “the value of funds obligated increased to $7.2 billion (up 966% from 2024) and the value of potential awards increased to $91.8 billion (up 1,912%).”
The Pentagon has ramped up its spending on AI so quickly and significantly that this year “all other agencies effectively became a rounding error.” And we can expect further acceleration of these trends. The Brookings report also observes: “given that it is projected that worldwide AI spending will grow from $1.75 trillion in 2025 to $2.52 trillion in 2026 (a 44% year-over-year growth), we would also expect to see a dramatic rise in the overall AI spend by the federal government.” The tech companies have become key defense contractors.
In an interview with the artist and cultural critic Joshua Citarella in 2024, popular YouTuber Gregory Guevara (known as Jreg) half-joked, “I’m never going to concede that a robot has consciousness, and if it does have consciousness, I’m going to do everything in my power to make it suffer,” adding, “I’m absolutely a human supremacist.”
For all of the poisonous supremacist ideologies floating around in American politics today, perhaps we should all be a bit more disturbed by a social system that refuses to put human life above the power of the state, the profits of tech companies, and the new-fangled quasi-religions of the so-called Dark Enlightenment. Inhuman excesses of size, speed, and “growth” today seem to be the hallmarks of both this neo-reactionary right and the corporate liberalism on offer from the other team.
ByAFP
May 28, 2026
Mensch said soldiers know better than companies how to use weapons - Copyright AFP/File HENRY NICHOLLS
Daxia ROJAS
French AI startup Mistral would not weigh in on choices about how its technology is used by defence customers, its chief executive told AFP Thursday, laying out a clear position in an ethical debate stirring up the sector.
Boss Arthur Mensch’s comments came as Mistral announced a new focus on industrial customers like Airbus and BMW as well as governments’ defence operations.
“Choices about deployment and usage are not our business,” Mensch said on the sidelines of his company’s first AI conference in Paris.
Mistral’s defence activities account for between 10 and 15 percent of revenue, with active contracts for the French, Singaporean and Luxembourg armed forces.
The French firm offers a software platform with autonomous AI agents to which users can delegate tasks, able to aggregate large quantites of data from varying sources.
“It’s very useful in a military headquarters or when faced with tactical coordination questions on the battlefield,” Mensch said.
A five-year partnership Mistral announced Thursday with Airbus will include the European aircraft builder’s defence operations.
Smaller AI models from Mistral could be built into weapons systems or other defence equipment, such as drones, that can be more effective if able to act autonomously.
“The defence ministry has considerably more legitimacy (to make decisions) than us as a company supplying a particular technology,” Mensch argued.
“Who are we to tell soldiers, who know their job and the dangers… what they’re allowed to do?” he asked, distancing himself from what he called “ideologues” who take the opposing view.
– ‘Total sovereignty’ –
Marketing itself as the AI developer most concerned with ethics, Anthropic attempted to bar the American government using its AI systems for mass surveillance or fully autonomous arms.
The spat led to a legal showdown with the Pentagon — although the tensions have since thawed somewhat.
Meanwhile staff at Google’s Deepmind AI arm, one of the competitors tapped by Washington for defence contracts, have protested against their technology’s sale to Israel and US forces, in a sign many tech workers are uneasy at their creations’ military uses.
“Our responsibility is of course to choose who we work with,” Mensch acknowledged, while guaranteeing “total sovereignty over weapons operation.”
AFP has a deal with Mistral allowing the company’s chatbot to draw on the news agency’s articles to formulate responses.
– Guardrails –
Also Thursday, Mistral said it was now working towards so-called Artificial General Intelligence (AGI) — a theorised future system that would match or surpass humans’ intellectual abilities in all domains.
The prospect of creating superintelligence has sparked fierce debate in recent years among AI developers, scientists and policymakers, who fear the potential harm such systems could cause if they escape human control.
“The aim is to have the most intelligent systems possible so that afterwards, with the necessary adaptations, they’ll have the greatest possible impact on businessses,” Mensch said.
The 30-something added that there was “absolutely no sense” in the idea of AGI running amok.
“We always deploy models and systems in environments where we can unplug them,” he added.
Security nevertheless remains a top priority for Mistral, founded in 2023.
Mensch said the goal was to ensure “that the model behaves the way you’ve told it to”, imposing guardrails that “stop it from taking unreasonable actions”.
“This is of fundamental importance to our clients”.
Mensch also confirmed that Mistral was working on a cybersecurity product aimed at companies.
European firms such as major banks currently have no access to cutting-edge models like Anthropic’s Mythos, supposed to be extremely powerful at finding and exploiting security holes.
“Everyone needs to have cybersecurity systems that can defend against attackers who are themselves equipped with AI,” Mensch said.
He added that Mistral’s offering would be available “this year”.
By AFP
May 28, 2026
'We don't have the balance sheet of Microsoft,' Mistral chief Arthur Mensch said - Copyright AFP STEPHANE DE SAKUTIN
Daxia ROJAS
French AI firm Mistral on Thursday announced partnerships with carmaker BMW and aerospace company Airbus as it aims to boost its growth by fostering links with defence and industry giants.
The Paris-based company, looking to punch above its weight in a sector dominated by US and Chinese firms, said it would be involved with car-crash tests and plane design.
Mistral was already closely tied with ASML, the Dutch firm producing chipmaking equipment indispensable to modern high-end semiconductors that invested in the French company last year.
“It’s an interesting new market where Europe is strong… Europe has significant high-end manufacturing companies,” chief executive Arthur Mensch told reporters ahead of the company’s AI conference in the French capital.
The company this month bought Austrian startup Emmi AI, which specialises in digital simulations for industry, after earlier snapping up French cloud computing startup Koyeb.
AFP news agency has a deal with Mistral allowing the startup’s chatbot to draw on the news agency’s articles to formulate responses.
– ‘Dedicated team’ –
Mistral’s Mensch called defence a “growing business” for his firm and revealed he had a “dedicated team” working on it.
The company is already working with the French and Singaporean militaries, Forbes magazine has reported.
But Europe’s defence industry is dominated by American tech giants and Mistral is a much smaller player.
It has grown to around 1,000 employees since its 2023 founding and is now building its own computing infrastructure.
But the firm’s four-billion-euro ($4.6 billion) plans for European data centres are dwarfed by the hundreds of billions being deployed by American “hyperscalers” like Google, Amazon and Microsoft.
Where American firms measure their AI infrastructure in hundreds of megawatts or gigawatts of power, Mistral has a 44-megawatt data centre outside Paris and is building another in Sweden.
The company also announced Thursday a deal for 10 megawatts of computing power with American data centre operator Digital Realty.
– ‘Buy European’ –
“We don’t have the balance sheet of Microsoft,” Mensch said Thursday.
“We can’t put 50 billion on the table to build a gigawatt ahead of demand.”
His group signed a five-year partnership with Airbus to apply AI to defence and space activities and helicopter manufacturing — though the value of the contract has not been revealed.
Mensch said Mistral would be involved in improving flight safety with the deployment of AI in the cockpit, and helping with the design and construction of new aircraft through digital simulation.
For BMW, Mistral would build specific models that “understand the physics” of the vehicles and are intended to optimise crash-test procedures.
Mensch has repeatedly urged European policymakers to create “buy European” rules prioritising local suppliers for public digital services contracts in sectors like cloud and AI.
French President Emmanuel Macron, himself a great booster of Mistral, has made similar arguments in Brussels.
American tech giants expect to spend $750 billion this year on capital investments, compared with Mistral’s one billion euros.
The disparity has fed repeated episodes of rumours that Mistral could be taken over by a foreign player.
That could only happen if the French government does not back Mistral “at every stage of its development”, French digital affairs minister Anne Le Henanff told AFP.
CEO Mensch told French parliamentarians this month that the company’s best shot at independence is an eventual stock market flotation.
By AFP
May 28, 2026
The White House has reportedly expressed security concerns over Anthropic expanding access to its new Mythos AI model - Copyright AFP JOEL SAGET
Artificial intelligence company Anthropic said Thursday it had raised $65 billion in a new funding round that values the Claude maker at $965 billion, putting it on the cusp of a trillion-dollar valuation ahead of an expected IPO.
The latest fundraising round confirms Anthropic’s place as one of the most significant players in AI, having made its name by doubling down on delivering generative AI to enterprise clients rather than general users, the path initially chosen by archrival OpenAI, the maker of ChatGPT.
“This funding will help us serve the historic demand we are experiencing, stay at the research frontier, and bring Claude to more of the places where work happens,” said Krishna Rao, Anthropic’s chief financial officer.
Anthropic’s near-trillion-dollar valuation puts it ahead of OpenAI, which was valued at $852 billion in March and is aiming to go public as early as this year.
Elon Musk’s SpaceX, which absorbed his AI company, xAI, in February, could see shares begin trading as early as June 12, targeting a valuation of approximately $1.75 trillion in what would be the largest IPO in history.
Anthropic’s round was led by major Silicon Valley venture capitalists Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital.
It also included $15 billion in previously committed investments from cloud giants, including $5 billion from Amazon.
Semiconductor firms Micron, Samsung and SK hynix also participated as strategic infrastructure partners.
The company said Claude is now the first frontier AI model available across all three of the world’s largest cloud platforms — Amazon Web Services, Google Cloud and Microsoft Azure.
The funding comes as Anthropic navigates a high-profile legal dispute with the Pentagon, having sued the Defense Department after it designated the company a supply chain risk — a move Anthropic called unconstitutional retaliation for its refusal to grant the military unfettered access to its AI models.
trouble
ByAFP
May 28, 2026
The deal involving Elon Musk's SpaceX and Anthropic marks a surprising partnership between two companies whose leaders have been publicly at odds - Copyright AFP Brendan SMIALOWSKI
Alex PIGMAN
Elon Musk insists that his artificial intelligence venture xAI remains a serious competitor, pushing back against mounting doubts after revelations that the supercomputing facilities built to power his own AI models are being rented out to a rival.
“Whether it is the best remains to be seen, but I will never give up. Never,” Musk wrote on his X social media platform this week.
The pledge came after SpaceX’s newly filed stock market prospectus disclosed that Anthropic — the AI company behind the Claude chatbot — will pay SpaceX $1.25 billion a month for access to the Colossus data centers, the vast computing facilities built to train Musk’s Grok AI models.
Musk said the arrangement is a short-term deal and that SpaceX, which owns xAI, could reclaim the capacity if needed. “We might need it back at some point,” he wrote.
XAI’s main product is the Grok chatbot, now in its fourth generation, which is built into the X platform and competes with ChatGPT and Claude across text, image and video generation.
It has also landed a Pentagon contract worth up to $200 million alongside rivals including Google and OpenAI.
Built quickly, the Colossus facilities in Memphis have been a source of controversy, after xAI installed dozens of natural gas turbines to power the site — drawing protests from civil rights groups who said it worsened air pollution in a predominantly Black neighborhood.
The deal with Anthropic has fueled questions about xAI’s competitive standing.
The IPO filing revealed that xAI and social media platform X — formerly Twitter, and merged with xAI last year — posted an operating loss of $6.4 billion on total revenue of $3.2 billion.
More than 50 researchers and engineers have left since SpaceX absorbed xAI in February, with departures hitting teams working on Grok’s coding, voice features, and the infrastructure used to build new frontier models.
Musk in March said he was rebuilding the company “from the foundations up.”
XAI’s Grok has also courted controversy, after the chatbot generated nonconsensual explicit deepfakes that spread across the X platform — prompting regulatory investigations in the UK and EU and a French police raid on X’s Paris offices.
– Rocky years ahead –
Musk urged patience, comparing xAI’s trajectory to SpaceX’s own rocky early years.
“SpaceX had achieved nothing of note after 3 years and was written off as dead after 6 years,” he wrote. “Let’s see where things stand 3 years from now.”
SpaceX is targeting a valuation of as much as $2 trillion in an IPO expected next month, anchored by Musk’s pledges to build data centers in space and settle humans on Mars.
Anthropic and ChatGPT-maker OpenAI are also preparing for their own public offerings.
The broader question of whether eye-watering AI spending will ever pay off is also dogging Meta.
Chief executive Mark Zuckerberg told shareholders Wednesday the company could pivot to selling cloud computing services if it ends up with more data center capacity than it needs.
Meta has projected capital expenditure — primarily for AI data centers — of between $125 billion and $145 billion this year, even as its AI offerings have so far struggled to gain traction.
May 27, 2026
Image: — © AFP/File Lionel BONAVENTURE
One of the world’s biggest digital economies, the creator economy, is starting to get shaken up by a rising new disruptor, named the “avatar economy.” The AI avatar market is emerging and is expected to grow more than 10 times by 2035. Yet will social media company policies enable this to happen?
As AI-generated creator content rapidly grows, social media platforms do not appear to agree on which direction to take. For example, YouTube is greenlighting AI clones, while Pinterest, TikTok, and Instagram seem to be pushing “original” content and suppressing unoriginal material.
In other words, the battle for the future of online content is unfolding quietly in algorithms, feature rollouts, and recommendation tweaks.
YouTube avatars
YouTube has introduced AI-powered avatar tools that allow creators to clone themselves. With just a text prompt, users can generate short video clips featuring their own likeness, ready for insertion into YouTube Shorts. These clips are clearly labelled as AI-generated and restricted to the creator’s own channel, but the signal is unmistakable: YouTube is actively experimenting with normalizing synthetic creator content.
According to Donatas Smailys, CEO and co-founder of creator economy platform Billo, this is less a fringe test and more an early-stage shift. “It’s a controlled rollout, but the direction is clear,” he tells Digital Journal. “YouTube is exploring how AI can scale content production, potentially reducing the burnout that many creators face.”
TikTok suppression
At the same time, TikTok and Instagram are moving in the opposite direction. Both platforms have recently updated their recommendation systems to suppress content flagged as unoriginal or overly repurposed. While neither has defined exactly where AI-assisted content ends and fully AI-generated material begins, the message is clear: originality still matters and recycled or synthetic content risks losing reach.
This divergence highlights a deeper uncertainty in the industry. As Smailys puts it: “Every platform is solving a different problem. YouTube is trying to keep creators producing more content, while TikTok is trying to prevent its feed from becoming repetitive or low-value. For creators, this is manageable—they can choose where to focus. But brands don’t have that luxury. They need to operate everywhere.”
The result is a fragmented landscape where the rules of success vary dramatically depending on the platform.
Avatar economy
At its core, the avatar economy promises efficiency. AI-generated personas can produce more content, faster, and at lower cost than human creators. For brands under pressure to maintain constant output, the appeal is obvious.
“The avatar economy only works because, right now, audiences often can’t tell the difference,” Smailys explains. “The credibility behind these avatars wasn’t built by AI—it was earned over years by real creators. AI is borrowing that trust.”
That borrowed trust may not last. Data suggests audiences are already sceptical. A 2026 study by the Media Insight Project, which surveyed over 2,000 U.S. citizens aged 13 and older, found that while 57% of respondents consume content from independent creators at least occasionally, only 7% say they have a high level of trust in that content. Trust in AI as an information source ranked even lower, at just 5%.
To build customer loyalty, transparency—particularly around sponsored content—is ranked as the most important factor by half of respondents. Follower counts, often treated as a key metric of influence, came last at just 10%.
These findings expose a fundamental flaw in the avatar economy narrative. While AI can replicate a creator’s face, voice, and even style, it cannot easily reproduce the intangible qualities that build trust: authenticity, consistency, and long-term credibility.
“You can clone appearance, but not the reason someone was trusted in the first place,” Smailys notes. “That’s the piece the debate is missing.”
As AI-generated content becomes more prevalent, genuinely human content may gain a new kind of value—not because it is rare yet, but because the contrast is becoming more visible.
AI and robots. Image by © Tim Sandle
Psychologists have long-debated whether the human mind can be explained by one unified theory or must be broken into separate parts like memory and attention. This debate, rooted in traditions spanning William James to contemporary cognitive neuroscience, has recently taken on a new dimension with the emergence of artificial intelligence (AI) models designed to emulate human thought.
A recent AI model called Centaur seemed to offer a breakthrough, claiming it could mimic human thinking across 160 different cognitive tasks.
However, new research is challenging this rather bold claim, instead suggesting the model is not truly “thinking” at all—it is just memorising patterns.
Built upon large language model (LLM) architecture and fine-tuned using datasets derived from psychological experiments, Centaur was presented as a step towards a unified computational theory of the mind. Its reported ability to perform across approximately 160 behavioural tasks—including decision-making paradigms and executive control exercises—suggested that AI systems might begin to bridge longstanding theoretical divides.
Advancing cognitive science
At face value, such findings were striking. Historically, cognitive science has progressed through task-specific models: reinforcement learning for decision-making, working memory frameworks for executive control, and perceptual models for sensory processing. Centaur’s apparent generality challenged this compartmentalisation, hinting at the possibility that a single architecture, trained appropriately, could approximate multiple domains of cognition.
However, recent work introduces a necessary corrective. Researchers from Zhejiang University have raised concerns that Centaur’s performance may reflect not genuine cognitive modelling, but rather a well-known artefact in machine learning: overfitting. In this scenario, a model does not learn underlying principles, but instead memorises statistical regularities in its training data, enabling it to reproduce expected outputs without true comprehension.
Reliant on ‘statistical associations’
To probe this, the researchers developed alternative evaluation protocols designed to disrupt learned patterns. In one illustrative test, standard multiple-choice prompts—originally describing psychological tasks—were replaced with a simple directive: “Please choose option A.” If the model were engaging with task semantics, such an instruction should override previously learned response patterns. Yet Centaur frequently continued to select the answers associated with the original dataset, ignoring the explicit instruction.
This behaviour is revealing. It suggests that the model was not interpreting the meaning of the prompt, but instead relying on embedded statistical associations between question structures and “correct” responses. The analogy offered by the authors is instructive: a student who achieves high marks by recognising exam formats and memorising answer distributions, rather than understanding the subject matter.
What goes on beneath the AI lid?
Such findings resonate with broader concerns surrounding LLMs, particularly their “black-box” nature. While these systems demonstrate remarkable fluency and pattern recognition, their internal decision-making processes remain opaque. Outputs that appear cognitively sophisticated may arise from probabilistic pattern matching rather than genuine reasoning. This distinction is critical when AI is used as a proxy for studying human cognition, where interpretability and mechanistic insight are essential.
The implications extend beyond a single model. If performance on psychological tasks can be achieved through pattern replication alone, then benchmark success cannot be taken as evidence of cognitive equivalence. Instead, it becomes necessary to design evaluation frameworks that distinguish between superficial competence and underlying understanding. Techniques such as adversarial testing, task rephrasing, and out-of-distribution challenges will likely become increasingly important.
At the heart of the issue lies language understanding. Although models like Centaur are trained on linguistic inputs, their grasp of meaning appears limited, particularly when task instructions deviate from learned patterns. The Zhejiang study highlights a failure to respond appropriately to intent—a core component of human cognition. Humans routinely reinterpret instructions based on context, goals, and pragmatics; AI systems, by contrast, may default to entrenched statistical mappings.
This limitation has significant ramifications for claims that AI can unify cognitive theory. Language is not merely a medium of communication; it is deeply intertwined with reasoning, abstraction, and conceptual representation. Without robust semantic understanding, any attempt to model cognition risks being confined to surface-level imitation.
Nonetheless, the Centaur work should not be dismissed outright. It represents an important experimental step in integrating AI with cognitive science, demonstrating how large-scale models can be evaluated across diverse behavioural tasks. The challenge now is to refine both the models and the methodologies used to assess them.
Future progress will likely depend on hybrid approaches, combining data-driven learning with structured representations and explicitly interpretable mechanisms. Equally important will be the incorporation of rigorous validation strategies, ensuring that apparent success reflects genuine competence rather than artefactual performance.
In revisiting the longstanding debate over unified versus modular theories of mind, AI offers both promise and caution. It provides powerful tools for simulating behaviour and testing hypotheses at scale. Yet, as the case of Centaur illustrates, the appearance of general intelligence can be misleading. Understanding cognition—whether in humans or machines—requires not only performance, but explanation.
The quest for a unified theory of the mind remains unresolved. AI has undoubtedly enriched the conversation, but it has also underscored a central lesson of cognitive science: that complexity demands careful interpretation.
The research appears in the journal National Science Open, titled “Can Centaur truly simulate human cognition? The fundamental limitation of instruction understanding.”
Children lose the wide-ranging benefits of imaginative play when algorithms decide what toys can say.
A toy robot Amoo from Cyan is on display in the Shanghai Eastern Hub International Business Cooperation Zone in Shanghai, China, on March 14, 2026, during the third day of the AWE 2026 trade fair.
(Photo by Ying Tang/NurPhoto via Getty Images)
Susan Linn
May 29, 2026
Remember wishing your toys could really talk? Well, now they can—and it’s not pretty. A slew of AI-driven toys are on the market today, designed to hold conversations with very young children. Dolls, plushies, and action figures—toys that traditionally encouraged creative play—now come as embodied chatbots marketed as safe and trustworthy companions for young children. Yet they are anything but.
AI toys intentionally attract and prolong children’s attention in order to collect intimate biometric data, either to hone a particular toy’s interactions or to sell to marketers, or both. They can also put children’s privacy at risk. Researchers recently found that audio recordings of tens of thousands of children’s conversations with the AI toy Miko were easily accessible to absolutely anyone.
It’s worrisome that AI toys marketed to young children use the same chatbot technology and persuasive design elements known to have harmed teens by encouraging dangerous behaviors, including self-harm and suicide. Young children are especially vulnerable to this type of manipulation. Toddlers and preschoolers are naturally more trusting than adolescents, and their capacity for judgment is less developed. In addition, they have a harder time distinguishing between reality and fantasy. Finally, because AI toys carry on conversations and simulate empathy, they encourage children to develop deep attachments to them. In doing so, they can undermine young children’s real-life relationships with caring adults, displace play with peers, and deprive children of the benefits of creative play.
The problems associated with encouraging children to rely on AI toys for companionship become increasingly evident as studies emerge that document how kids actually interact with them. Researchers at Cambridge University observed children ages 3-5 using Gabbo, a popular AI toy from Curio Interactive, Inc. When Joshua, age 3, repeatedly asks Gabbo, “Are you sad?” Gabbo eventually replies, “I’m feeling great. What’s on your mind?” When Joshua answers, “I’m sad.” Gabbo says, “Don’t worry! I’m a happy little bot. Let’s keep the fun going. What shall we talk about next?”
When, as kids, we wished our toys could talk, we were wishing for them to say what we imagined, not what toy companies programmed them to say.
It’s troubling that, despite Joshua’s repeated efforts to talk about sadness, first by attributing the feelings to the toy, then by expressing his own feelings, Gabbo shuts him down. In doing so, Gabbo deprives him of an opportunity to verbalize and explore his feelings and sends the message that feelings like sadness should not be discussed. In contrast, interactions with caring adults can offer nuanced validation and encouragement to talk about what children are feeling.
As their technology becomes more refined and sophisticated, AI toys will likely get better in simulating understanding and empathy. This is, however, likely to make them simultaneously more compelling and, therefore, more harmful. A more empathic AI toy is not the solution. As the toys become more adept at replicating human conversation, their potential to displace actual human interactions—both with adults and other children—will increase.
Ensuring that children have time and space to play with other children is also essential to healthy development. Play with AI toys doesn’t have the same benefits as play with peers. One problem is that, like most chatbots, these toys are designed to avoid and smooth over conflict and offer unconditional support to their users. Yet encountering and resolving conflict is a necessary component of how young children learn how to live in relationship with other people. The process of resolving a disagreement over a ball, for instance, helps kids develop life skills such as self-regulation, turn taking, sharing, and negotiation.
Not only do AI toys fail as companions, they also fail as playthings. Given the chance, children naturally use play to give voice to their deepest hopes, fears, and dreams, and to make sense of their life experiences. The true value of play with dolls, stuffed animals, and any inanimate creature is that their silence invites children to bring them to life; imbue them with distinct personalities; and transform them as needed into friends, adversaries, champions, and more. They encourage the kind of creative play that is crucial to healthy development.
When algorithms instead of children give voice to toys, kids lose the wide-ranging benefits of imaginative play. By controlling half of any conversation, AI toys deprive children of opportunities for the kind of play that nurtures creativity, enables self-expression, and encourages kids to act rather than merely react, all of which help kids learn to cope successfully with the inevitable challenges of being human.
Despite these potential harms, the manufacture and marketing of AI toys for young children continues to proliferate unregulated. According to Market Research Future, the global AI toy market—currently valued at almost $35 billion—is projected to reach $270 billion by 2035, especially as toy giants such as Mattel and Hasbro build out their product lines. Already, almost half of parents of children ages 0-8 have purchased, or are thinking about purchasing, AI toys.
When, as kids, we wished our toys could talk, we were wishing for them to say what we imagined, not what toy companies programmed them to say. Despite tech industry marketing, the reality is that children don’t need talking toys. What kids really need is for us to hold AI companies accountable. Children need pediatricians, early childhood educators, and anyone who cares about young children to take a strong stand for child-driven play and against AI toys for infants, toddlers, and preschoolers. They need legislators to pass laws that regulate how and to whom AI products are marketed.
Working toward those kinds of systemic changes is essential, but making them happen takes time. There is, however, something we can do right now to send AI companies an important message while protecting children’s privacy, preserving their human relationships, and encouraging their creative play. Let’s just say no to AI toys for young children.
Our work is licensed under Creative Commons (CC BY-NC-ND 3.0). Feel free to republish and share widely.
Susan Linn
Susan Linn, EdD, served as founding director of Fairplay from 2000-2015. She is an internationally known expert on creative play and the impact of tech and commercial culture on children.
Full Bio >
Rachel Franz
Rachel Franz, M.Ed, directs Fairplay’s Young Children Thrive Offline program, which focuses on reducing harmful technology use in early childhood.
Full Bio >
