Bloomberg data shows AI job losses are reaching white-collar roles

Fresh US labour data suggests the first measurable job losses in occupations most exposed to AI are no longer hypothetical. A Bloomberg analysis of Bureau of Labor Statistics data found employment across 18 occupations judged highly exposed to AI — covering roughly 10 million workers — fell 0.2 per cent in the year to May 2025, even as overall US employment rose 0.8 per cent. That gap does not prove chatbots dismissed every worker who left those categories. What it signals is a shift in the burden of proof. For software teams, outsourcers and policymakers who have treated AI disruption as a medium-term risk, the question is no longer whether the data will eventually turn up. It is what the data is already counting.

The roles in question are not warehouse pickers or call-centre workers. They are the clerical and knowledge-work jobs that vendors spent the past two years describing as prime candidates for copilot tools: administrative support, basic analysis, documentation, customer handling and entry-level research. That same cohort shows up in the Washington Post’s interactive breakdown of exposed occupations, while a new Anthropic labour-market paper found young workers have been finding it harder to move into AI-exposed jobs — the job-finding rate into those roles is down 14 per cent since 2022. The pressure may be visible at the hiring margin before it appears in redundancy announcements.

A two-tenths of a percentage point decline sounds small. But it is moving in the opposite direction from the wider labour market, which is still adding jobs. It also lands after months of executives talking about shipping more output with flatter teams — in coding, support, marketing operations and back-office functions especially. A toolset sold as a productivity multiplier is now coinciding with weaker employment in the occupations economists and model builders tagged as most exposed. The overlap remains noisy. The exposure models have known blind spots, and a CEPR VoxEU note on AI occupational-exposure measurement warned that the way researchers construct those yardsticks affects what they appear to find. But noisy measurement cuts both ways: when a signal is visible through an early, imperfect proxy, employers and regulators can no longer dismiss it as a thought experiment.

The sceptical case deserves a hearing. White-collar employers spent the past two years trimming after a pandemic hiring surge. Higher interest rates and slower growth have forced margin protection. Some companies may be slotting AI into announcements about cuts they planned anyway. The supporting data sharpens the picture. A CBS News report on Challenger data said AI was cited as a driver of layoffs in April, accounting for 26 per cent of announced job cuts that month. Challenger’s Andy Challenger argued that whether or not individual roles are directly replaced by software, the budgets for those jobs are under pressure. Mark Muro of Brookings put a finer point on it in the Washington Post piece, warning that changes in white-collar work “may be out of sight and out of mind” because they surface first in slower hiring, smaller teams and tasks that quietly disappear — not in a single factory closure.

Australian employers should be paying attention.

Australia does not publish an equivalent data series linking AI exposure to occupational losses. But local tech and enterprise leaders are making the same investment case their US peers made first: more copilots, more automation, fewer repetitive tasks and leaner support functions. For Australian software businesses, the US numbers are an early signal about where labour-market effects are likely to surface before they become politically visible. They may appear first in graduate intakes, contract renewals, shared-services hiring and junior operations roles — not in dramatic all-staff memos. Australia’s pool of entry-level technical and administrative jobs is small to begin with, so a decision to absorb those tasks into tools or push them onto fewer senior staff can reshape career ladders quickly.

That sharpens the policy question. If AI pressure arrives first at the bottom of white-collar ladders, the workers with the least bargaining power are being asked to compete with systems sold as always-on, instantly scalable and cheap to deploy. Governments that want the productivity upside of AI adoption without a backlash from younger graduates and routine knowledge workers face awkward terrain. The response is unlikely to be a ban on AI tools. Disclosure requirements, training subsidies, transition support and better occupational data are more realistic — because the biggest risk is not a wave of sackings. It is a slow erosion of pathways into stable professional work. Once those pathways narrow, the damage is harder to reverse than a quarterly hiring freeze.

The AI industry’s early pitch contrasted generative tools with industrial automation. Manual labour, the argument went, would stay stubbornly human; software would help office workers move faster. The numbers tell a less comfortable story. Routine white-collar tasks were always easier to standardise, meter and route through software than the boardroom rhetoric admitted. If the first measurable employment weakness is showing up there, the debate has to move past the claim that AI merely augments. Augmentation can still cut headcount when one employee with better tools absorbs work that previously required two junior hires.

The numbers are too early for a recession call, and it would be premature to read them that way. But the labour-market argument has entered a different phase. Mainstream employment data, company layoff explanations and model-based research are pointing in the same direction for the first time: the most exposed office roles are under pressure. Australian readers do not need to assume the US path will map neatly onto local conditions. They do need to reckon with the fact that the story has moved from prediction to something harder to dismiss — a measurable shift in the jobs most exposed to the tools employers are racing to buy.