After months of warnings from technology leaders about artificial intelligence eliminating millions of jobs, fresh research suggests the actual impact on employment remains surprisingly modest. Anthropic economists Maxim Massenkoff and Peter McCrory have published a detailed analysis of labour market outcomes, and it paints a strikingly different picture from the alarmist rhetoric that has dominated public discourse.
The study, titled "Labor market impacts of AI: A new measure and early evidence," introduces what the researchers call "observed exposure" as a measurement tool. Rather than asking what AI could theoretically accomplish, observed exposure measures what AI is actually being used to do in real workplaces right now. That distinction turns out to matter enormously.
Workers in the most exposed occupations have not become unemployed at meaningfully higher rates than workers in jobs considered AI-proof, with the average change in the unemployment gap since ChatGPT's release described as "small and insignificant." For those relying on earlier predictions, this finding represents a striking reversal.
The centre-right case for fiscal caution has long emphasised the importance of evidence over speculation. Here, the data contradicts some of the more sweeping claims made by technology executives themselves. Anthropic CEO Dario Amodei, for instance, warned in January 2026 that AI could displace half of all entry-level white-collar jobs within one to five years. Yet the same company's economists found no broad-based employment collapse in the months since ChatGPT's release in late 2022.
The gap between theoretical capability and practical adoption proves vast.Actual AI adoption is just a fraction of what AI tools are feasibly capable of performing, with adoption measured using work-related usage data from Anthropic's Claude model. Computer programmers show the highest exposure, with AI theoretically covering 75% of programming tasks, yet no corresponding surge in programmer joblessness has materialised.
Still, the data contains a genuine warning worth taking seriously.Anthropic does find "suggestive evidence that hiring of younger workers" particularly ages 22 to 25 has "slowed in exposed occupations." This friction at the entry level deserves policy attention, even if it falls well short of the economic catastrophe some have predicted.
The legitimate counterargument gains weight here. Researchers at institutions including Yale and Stanford note that major technological disruptions often take years to fully materialise in employment statistics.The track record of past approaches to forecasting AI labour impact gives reason for humility, as prominent attempts to measure job vulnerability identified roughly a quarter of US jobs as vulnerable to offshoring, but a decade on, most of those jobs maintained healthy employment growth. History offers no guarantee that AI will follow a different pattern.
Anthropic CEO Dario Amodei last year said the technology could disrupt half of entry-level white-collar work. Yet Microsoft's AI chief Mustafa Suleyman predicted even more dramatic displacement. The divergence between these warnings and actual labour market outcomes raises fundamental questions about how seriously to treat future projections from the same sources.
The research also identifies which occupations face the most exposure.Computer programmers, customer service representatives, and financial analysts are among the most exposed. More broadly,the most exposed workers are more likely to be female, more educated, and higher-paid, with this wave hitting knowledge workers first. This distributional aspect matters; disruption concentrated among high-income workers creates different policy problems than dispersed, economy-wide unemployment.
What emerges from the evidence is neither the doomist scenario of mass joblessness nor the complacent view that AI poses no employment challenges. The honest assessment falls between these extremes. Real technologies do reshape labour markets. Adoption timelines remain profoundly uncertain. And the difference between theoretical capability and workplace reality may prove far larger and more persistent than either enthusiasts or pessimists have assumed.
For policymakers, this demands a pragmatic middle path. Invest in transition support and reskilling for workers in genuinely exposed sectors, particularly young entrants to fields showing hiring friction. Avoid sweeping restrictions on AI development based on speculative harm projections. And most importantly, insist on continuous measurement.The goal should be to establish an approach for measuring how AI is affecting employment, and to revisit these analyses periodically, helping to more reliably identify economic disruption than post-hoc analyses.
The case for sound economic management rests on facing facts rather than fears. On current evidence, those facts suggest AI's employment disruption may be real, measurable, and concentrated in particular sectors and age groups, but far more modest than the loudest voices have claimed. That conclusion does not excuse inaction, but it does argue for targeted, evidence-based responses rather than panic-driven policy.