In the first months of 2026, at least 73,200 tech workers were laid off across 95 companies, according to tracking site Layoffs.fyi. The list includes Amazon (16,000), Oracle (planning 20,000 to 30,000), Meta (multiple rounds totalling several thousand), Snap (1,000, representing 16 percent of its workforce), and Disney (1,000 under its new CEO). In several of these cases, companies named AI directly. Snap's CEO said advances in AI were enabling automation of repetitive tasks. Oracle explicitly linked its cuts to AI data-center expansion. The pattern is hard to ignore. It is also, according to the International Labour Organisation, hard to measure properly — and that distinction matters.
The ILO released a brief this week warning against treating AI exposure indicators as direct forecasts of employment losses. The organisation's point is technical but important: the metrics researchers use to identify jobs "at risk" from AI are based on static job descriptions and theoretical feasibility, not on observed adoption decisions, profitability calculations, or how actual labour markets respond over time. Higher-skilled roles in finance, computing, and business now show greater AI exposure than the routine manual work that earlier automation studies focused on. But exposure is not displacement. A job being technically automatable and a company choosing to automate it are different events, separated by cost, organisational change, and regulatory environment.
The gap between the two shows up in real data. LinkedIn, which has more visibility into hiring patterns than almost any other organisation, has been pushing back on the narrative that AI is driving the current hiring slowdown in tech. The company's position, based on its own platform data, is that broader macroeconomic conditions — interest rates, post-pandemic correction, reduced venture funding — explain more of the contraction than AI adoption does. That framing sits uncomfortably next to Snap's CEO explicitly citing AI automation as the reason for his headcount reduction, but both things can be true simultaneously: macro conditions set the ceiling on hiring, while AI determines where within that ceiling companies choose to cut.
What is harder to resolve is the direction of the trend. The ILO's warning about misusing exposure statistics is methodologically sound. But it risks being read as reassurance when none is warranted. The 73,000 documented layoffs are not statistical projections: they are people who no longer have jobs, at companies that in several cases gave AI as the reason. The question the ILO is raising is whether those numbers scale to the broader economy, and whether the theoretical exposure of millions of roles translates into actual displacement at the rate the most alarming forecasts suggest. That remains genuinely unknown.
What is clear is that the measurement problem is real and consequential. If the statistics used to track AI's employment impact are unreliable, then policymakers making decisions about retraining programmes, social safety nets, and labour regulation are working with bad instruments. The ILO's recommendation is to treat exposure indicators as early warning signals combined with broader economic data, rather than as forecasts. That is reasonable advice. It does not, however, make the warnings less urgent: it makes them harder to act on precisely, which is its own kind of problem.