The AI jobs data contradicts itself

A survey of nearly 6,000 senior executives across the United States, United Kingdom, Germany, and Australia, published by the National Bureau of Economic Research, found that 90% of firms report no impact on employment or productivity from AI so far. ¹ Sixty-nine per cent of the surveyed firms actively use the technology. Nine in ten see nothing happening.

The contradiction sits inside the forecasts. Executives at these firms predict a 0.7% employment decline over the next three years. Employees at the same companies predict a 0.5% increase. ² One group expects cuts. The other expects growth. They work in the same buildings, use the same tools, and hold irreconcilable views of what comes next.

During the 1990s offshoring wave, management planned relocations years before workers learned their roles would move overseas. Approximately 3.4 million US manufacturing jobs were lost between 1995 and 2005. Workers could not prepare because they did not know. The NBER data, spanning four countries with different labour market systems, suggests this gap is structural, not cultural . If executives act on private bearish forecasts without informing staff, displacement will arrive as a shock rather than a managed transition.

The European Parliament voted 101 to 9 on 26 March to delay the AI Act's high-risk employment rules from August 2026 to December 2027, with eight MEPs abstaining. ¹ The Digital Omnibus amendments also push AI systems covered by EU sectoral safety legislation to August 2028. Trilogue negotiations with the Council and Commission now begin, but the Council agreed its own position on 13 March. Both chambers want the delay.

The vote ratifies a proposal the Commission first floated earlier this year . What was then a suggestion is now a near-certainty. More significant than the timeline shift is a provision that attracted less attention. Employer obligations to ensure staff AI literacy have been removed from provider and deployer requirements. ² That responsibility now sits with the Commission and member states, which is to say it sits with no one in particular.

Europe's AI Act was the only binding framework that required employers to understand the AI tools they deploy against their own workforce. Companies operating in the EU now face no compliance obligation on high-risk employment AI until late 2027. When that date arrives, their staff will have no guaranteed right to understand what the systems do. The framing was timeline adjustment. The substance is policy retreat.

A joint study by the International Labour Organisation and the World Bank, published in March and covering 135 countries and two-thirds of global employment, found that one in four workers holds a job with some generative AI exposure. ¹ In the highest-risk category, 4.7% of women are exposed globally, compared with 2.4% of men. In high-income countries the gap widens: 9.6% of women versus 3.5% of men.

Clerical and administrative work is the exposure vector, roles historically concentrated among women. Advanced economies show 34% overall workforce exposure; low-income countries show 11%. ² Wealth correlates with AI penetration into clerical roles where women are concentrated. Wealth also correlates with the gender disparity.

This dimension has been absent from the displacement debate. CFO surveys, tech layoff trackers, and congressional proposals: none address gendered impact. The 502,000 roles the Atlanta Fed projects will be cut are not distributed evenly. Clerical and administrative roles most exposed to generative AI are held disproportionately by women, and no current policy framework accounts for it.

In 2025, New York State updated its Worker Adjustment and Retraining Notification Act to require companies to disclose AI's role in mass layoffs, becoming the first US jurisdiction to mandate such reporting. After nearly a year of operation, the results are in. ¹ Zero of 162 companies filing layoff notices attributed cuts to AI or technological automation. Those filings covered more than 28,300 workers, including staff at Amazon and Goldman Sachs.

Non-compliance currently carries a penalty of $500 per day. Proposed legislation would raise that to $10,000 per violation and strip companies of state grants and tax incentives for five years. That tougher bill has not advanced.

Silence on this scale is evidence, not absence. Harvard Business Review reported that only 2% of layoffs followed actual AI deployment . Oxford Economics called AI's layoff role "overstated" . Both relied on corporate claims taken at face value. New York's data shows those claims are legally shielded as well as reputationally incentivised. Companies that cut 28,300 jobs had the opportunity and the obligation to say whether AI played a role. Every one said no. Either AI genuinely drives none of the displacement in the nation's financial capital, or the disclosure framework is failing.

Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez introduced the AI Data Centre Moratorium Act on 25 March. ¹ The bill would ban all new AI data centre construction until Congress passes legislation addressing worker protection, consumer rights, civil rights, and environmental standards. It cites electricity costs rising nearly 7% last year, double the overall inflation rate, costing the average US household an extra $123 in 2025.

This is separate from Sanders' earlier robot tax proposal . He now operates on two tracks: one targeting AI's economic output through taxation, the other targeting its physical infrastructure through permitting. The moratorium has no path under a Republican-controlled Congress. It is a negotiating position, not legislation designed to pass. Its function is to define the left boundary of the debate and force centrist proposals to account for energy and environmental costs alongside labour displacement.

A bipartisan Coalition of 9 US senators wrote to the Department of Labour, the Bureau of Labour Statistics, and the Census Bureau in March, urging expanded data collection on AI's workforce effects. ¹ Senator Mark Warner and Senator Josh Hawley, who introduced the AI-Related Job Impacts Clarity Act last year , lead the Coalition. Seven additional signatories joined: Jim Banks, Maggie Hassan, John Hickenlooper, Mark Kelly, Tim Kaine, Mike Rounds, and Todd Young.

No bill has advanced. But the Coalition's growth from two sponsors to nine signatories, drawing from multiple committees, is the strongest signal yet that AI workforce accountability has a durable political centre. While Sanders targets infrastructure and taxation, this group targets measurement. Neither approach has produced a law. Federal agencies can act on the data request without new legislation, which may make it more consequential than either bill.

Microsoft reported first-quarter fiscal 2026 revenue of $77.7 billion, up 18% year on year and beating consensus of $75.6 billion. ¹ Azure cloud revenue grew 40%, the fastest rate in over a year. The company spent $34.9 billion in capital expenditure during the quarter and claims 900 million monthly active users of AI features, with 150 million on Copilot and 26 million on GitHub Copilot.

These numbers matter because the Big Five committed $650–690 billion to AI infrastructure this year , and Barclays forecast that the spend would sharply reduce free cash flow: Meta down 90%, Microsoft down 28% . Microsoft's results are the first to land, and they beat expectations. The counter-case has not collapsed. Microsoft converts AI spending to cloud revenue more directly than any peer. Meta reports on 29 April; its business model converts AI capex to advertising efficiency, a slower return. One quarter of good numbers from Redmond does not validate $690 billion in collective spending. But it makes the bear case harder to sustain without Meta's data.

Atlassian announced on 24 March that departing CTO Rajeev Rajan will be replaced by two AI-specialised chief technology officers: Taroon Mandhana (CTO of Teamwork, overseeing the Rovo AI collaboration platform) and Vikram Rao (CTO of Enterprise and Chief Trust Officer). ¹ Sixty-three Washington State workers are among those affected by the company's broader 1,600-person layoff .

This is the first public example of a tier-one software company decomposing its top technical role specifically to accommodate AI's operational demands. One executive owns AI product development. The other owns enterprise trust and compliance. Companies cutting workers are simultaneously restructuring how they govern the technology replacing them.

Close Brothers, a UK banking group, announced 600 job cuts on 23 March as part of an £85 million cost-cutting programme, citing AI deployment and offshoring. ¹ The announcement came one day after a short-seller warned of a £1.2 billion hit from the car finance scandal.

AI and offshoring appear in the same sentence. When companies face financial pressure from multiple directions, AI becomes a convenient explanation for cuts that serve other purposes, precisely the pattern the "AI-washing" thesis predicted .

Epic Games cut more than 1,000 jobs, roughly 20% of its workforce, on 24 March. ¹ CEO Tim Sweeney explicitly denied AI played any role, pointing instead to declining Fortnite engagement and the company spending more than it earns. Severance includes a minimum of four months' base pay and six months of US healthcare coverage.

Sweeney's denial stands out. In a quarter where one in five tech layoffs cite AI , he went out of his way to reject the label. Not every cut is an AI story.

UK job vacancies fell to 721,000 in the three months to February, down 9.5% year on year, with declines across 15 of 18 industry sectors, the Office for National Statistics reported on 18 March. ¹ Unemployment rose to 5.2%. Nominal wage growth held at 3.8%, but in real terms workers gained just 0.4 to 0.5%.

Workers losing ground to inflation have less financial runway to absorb a job loss or fund retraining. This squeeze coincides with the UK Government's own projection that AI-direct jobs will surge, but only in professional roles.

In January, the UK Government published projections showing AI-direct employment rising from 158,000 jobs in 2024 to 3.9 million by 2035, with up to 9.7 million in AI-related occupations. ¹ Ninety per cent of net employment growth falls in professional and associate professional roles. Declines concentrate in administrative, secretarial, and skilled trades.

AI roles are growing, but the workers being displaced from clerical and administrative positions are not the ones who will fill them. The government's own data amounts to an admission that displacement will fall disproportionately on mid- and lower-tier workers, while new opportunities cluster among those who least need them.

The US Department of Labor reported initial jobless claims of 205,000 for the week ending 14 March, down 8,000 from the prior week. ¹ Insured unemployment fell to 1,819,000, the lowest level since May 2024.

The number sits awkwardly alongside nonfarm payrolls that fell by 92,000 in February and tech layoffs now past 59,000 for the year. Claims data captures workers filing for benefits; payroll data captures jobs created and lost. The two metrics are telling different stories about the same economy.