'Sugarcoating the Risk': Anthropic Boss Warned of 20% Job Loss, But His Own Chief Economist Disagrees

Anthropic chief economist Peter McCrory published an analysis on 24 July finding no evidence of mass AI-driven unemployment in current US labour data, directly contradicting CEO Dario Amodei's warnings that AI could push unemployment to 10-20% within one to five years. McCrory cited a June unemployment rate of 4.2%, near 'full employment' by the Federal Reserve's measure, and found 'no relative deterioration' in occupations most exposed to AI tools like Claude.

'Sugarcoating the Risk': Anthropic Boss Warned of 20% Job Loss, But His Own Chief Economist Disagrees New research from Anthropic reveals no evidence of mass unemployment, setting up an internal corporate clash over the true economic impact of artificial intelligence. Anthropic chief economist Peter McCrory has publicly shattered corporate orthodoxy by rejecting CEO Dario Amodei's apocalyptic warnings of mass artificial intelligence-driven unemployment. As the global AI job loss debate intensifies across digital workspaces, an exhaustive 18-month empirical examination of the US labour market AI impact reveals a stunning contradiction between top-down executive panic and grounded federal statistics. McCrory argued in a new analysis published on 24 July that current US labour data show no sign of mass unemployment driven by AI. While leadership warns of catastrophic structural displacement affecting up to twenty percent of the working population, real-world macroeconomic numbers tell a vastly different, more measured narrative. Peter McCrory's AI Findings Challenge Job-Loss Narrative McCrory laid out his findings in a lengthy essay on X, setting out what is essentially Anthropic's in-house economic view of AI's labour impact so far. He began with the headline indicators. In June, the US unemployment rate stood at 4.2%, which the Federal Reserve regards as roughly 'full employment'. Job openings were in the same ballpark as the number of unemployed workers. Employment among so‑called prime-age workers was close to multi‑decade highs. In other words, if AI were already triggering the kind of jobs shock some fear, it was not visible in the national statistics. McCrory did not stop there. He then zeroed in on the occupations where Claude is most heavily used to automate tasks, and compared unemployment rates in those jobs with roles that are far less exposed to AI tools. His conclusion was blunt: he found 'no relative deterioration' in the more AI-exposed group. 'I don't expect unemployment to be noticeably higher a year from now at least not because of AI,' he wrote. Nothing in the essay suggests McCrory is complacent. But he is clearly pushing back, gently but firmly, against the idea that the jobs crisis is already underway. A big part of his explanation lies in what he calls AI's 'stubbornly jagged' capability profile, borrowing a phrase from Wharton professor Ethan Mollick. No single occupation tracked by the US Labour Department has all of its tasks fully handled by Claude. When McCrory looked at how workers actually use the system day to day, he saw something more prosaic than full automation. People were folding Claude into their workflows to draft, refine and iterate, not handing over entire jobs wholesale. Peter McCrory Versus Dario Amodei on AI's Jobs Future That measured picture sits uneasily alongside Amodei's public statements. To recall, in May 2025 he said that AI could erase up to half of all entry-level white-collar roles and push US unemployment into the 10% to 20% range within one to five years. He accused companies and policymakers of 'sugarcoating the risk' and urged them to stop. In January 2026, Amodei went further, describing AI as a 'general labour substitute' and predicting that work would be pushed upwards into fewer, higher‑skill positions, potentially leaving many workers either permanently sidelined or trapped on very low wages. By June that year, he was arguing for universal basic income and wage insurance, and warning that significant job loss might be 'an intrinsic property of the technology.' McCrory's analysis does not directly refute those dire projections. The honest reading is more awkward. His data show that the doomsday scenario has not arrived yet, at least not in the aggregate statistics that capture the broad labour market. Both men are looking nervously at the same group: early-career workers in highly AI-exposed roles. Their split is over timing and scale how deeply and how fast the knife will cut. There are, to be fair, rumblings in the background. McCrory acknowledges that hiring has already softened for younger workers in roles with high AI exposure https://www.ibtimes.co.uk/ai-automation-skills-gap-mit-expert-warning-1809796 over the past year. Separate Stanford research he cites has described these workers as 'canaries in the coal mine'. The Bureau of Labour Statistics is also projecting slower employment growth through 2034 in precisely the kinds of jobs that tools like Claude are best suited to, including technical writing, data entry and customer support. Where Peter McCrory Thinks the Real AI Shock May Land McCrory's essay raises an uncomfortable question that tends to get lost when the conversation is dominated by raw job counts. If AI is already lifting productivity for some workers without blowing a hole in employment, who is actually benefiting from that extra output? The answer he sketches is lopsided. Companies whose staff are fluent with AI can produce more without hiring more, and those gains show up quickly on the earnings line. But the upside is concentrated in particular sectors, in larger or more technologically aggressive firms, and among workers already in higher‑skill, higher‑income roles. Everyone else mostly jogs in place. The risk is obvious enough. If the productivity boost remains bottled up among AI power users, the income gap between those workers and the rest of the labour force widens further. Since lower and middle‑income households tend to spend a greater share of their earnings on day‑to‑day consumption, persistent wage divergence at the bottom eventually drags on consumer spending growth. That is not an overnight crash, but a long, grinding economic problem that compounds quietly over time. Crucially, McCrory argues that the apparent delay in any full-blown AI labour shock buys time not comfort. If job disruption is indeed real but still building beneath the surface, there is a window for governments and employers to move on retraining, education and shoring up safety nets before unemployment climbs. By contrasting internal Claude usage metrics with official Bureau of Labour Statistics figures, McCrory exposes the vast chasm between theoretical automation fears and actual workplace realities, forcing global policymakers and industry leaders to rethink how digital intelligence truly transforms modern employment before algorithmic panic completely outpaces economic truth. © Copyright IBTimes 2025. All rights reserved.