AI Isn't Cutting Jobs Yet. It's Blocking Young Talent from Entering.

The unemployment figures aren't showing the disruption yet. But look at who's getting hired, and a different picture emerges. Research from Anthropic reveals that whilst AI hasn't triggered mass job losses in the two years since ChatGPT's arrival, young workers trying to break into AI-exposed professions are finding doors increasingly closed to them.

The job-finding rate for workers aged 22 to 25 in occupations highly exposed to large language models has dropped by roughly 14 per cent compared with pre-2022 levels, according to the study. The company's researchers flag the finding as 'only marginally statistically significant'—a qualification that matters when interpreting what could be an early warning signal or merely statistical noise.

AI may not be eliminating jobs outright, but rather removing the bottom rungs of career ladders whilst leaving established workers largely untouched.

What makes this particularly striking is the demographic reality of who actually faces AI exposure. The narrative around automation typically conjures images of low-wage workers rendered obsolete by machines. The data tells a different story entirely.

Enjoying this article?

Get stories like this in your inbox every week.

Join the Community

The surprising profile of AI-exposed workers

Workers in occupations most exposed to AI tools tend to be older, female, highly educated and higher-paid than their counterparts in roles where AI plays little or no part. Computer programmers, customer service representatives and data entry workers sit at the top of the exposure rankings—precisely the sorts of white-collar positions that once represented stable middle-class employment.

Around 30 per cent of occupations show no meaningful AI coverage whatsoever, according to Anthropic's analysis. Cooks, mechanics, bartenders and lifeguards remain largely insulated, their work too physically grounded or context-dependent for current AI systems to replicate. The technology that was meant to democratise knowledge work may instead be creating a curious inversion: those with advanced degrees face greater exposure than those working with their hands.

This creates an awkward counterintuitive reality for both policymakers and workers trying to gauge where the risks actually lie. The most vulnerable aren't necessarily those at the bottom of the income distribution, at least not yet.

From theoretical capability to actual deployment

Anthropic's research introduces what it calls 'observed exposure'—a metric combining theoretical AI capabilities with real-world usage data from Claude, the company's chatbot. The distinction between what AI could do and what it actually does proves revealing. In computer and mathematical roles, current AI usage covers roughly a third of job tasks, despite models theoretically capable of assisting with far more.

That gap between potential and practice matters enormously for anyone trying to forecast AI's labour market impact. Legal constraints, technical integration challenges and the persistent need for human oversight all slow adoption. The pace of real-world deployment remains gradual, even as public discourse often assumes rapid transformation is already underway.

Yet gradual doesn't mean negligible, particularly for those trying to establish themselves professionally. Historical automation patterns suggest new technology often eliminates entry-level positions—think bank tellers or legal secretaries—long before it displaces experienced practitioners who've built institutional knowledge and client relationships. If AI follows that trajectory, the 14 per cent decline in young worker hiring could presage something more substantial.

Worth noting, too, that Anthropic has commercial incentives to frame its findings carefully. An AI company publishing research that downplays employment disruption faces an obvious conflict of interest. The study deserves credit for statistical candour about marginal significance, but readers should weigh the source when interpreting implications.

The making of a two-tier labour market

What emerges from this data is the outline of a potential two-tier labour market: one where established workers leverage AI to enhance productivity whilst remaining employed, and another where younger workers struggle to gain the initial experience that makes them valuable beyond what AI can provide.

If employers can use AI tools to handle tasks previously assigned to junior staff, the rationale for hiring and training those staff weakens considerably. Graduate employment pipelines, professional training programmes and traditional career progression models all assume a certain volume of entry-level positions exist.

Junior associates do the grunt work whilst learning from senior colleagues. Paralegals review documents whilst absorbing legal reasoning. Marketing assistants manage spreadsheets whilst developing strategic thinking. Remove enough of those positions, and you don't just affect this year's graduates—you undermine the entire mechanism through which professions reproduce themselves.

The implications for social mobility are particularly concerning. Professional careers have historically offered pathways for capable young people from ordinary backgrounds to advance economically. If those pathways narrow or close, the knowledge economy risks becoming increasingly hereditary—accessible primarily to those with family connections or resources to sustain them through extended unpaid internships or precarious freelancing.

Whether this pattern accelerates or proves transitory depends partly on how quickly AI capabilities advance and how thoroughly businesses integrate them. But it also depends on policy choices around training, employment protections and how societies structure professional development. The slow and expensive nature of current AI adoption means there's still time to shape outcomes, and new tools are being developed to help warn when jobs might be lost to AI. The research suggests we're still early enough to shape outcomes rather than simply react to them, assuming policymakers recognise the pattern before it calcifies into permanent disadvantage.

More articles by David Adams