The early read is calm, with one catch. Across the state, the researchers found no sign of a surge in AI-driven layoffs. The share of claims coming from AI-exposed workers has not risen meaningfully since ChatGPT arrived in late 2022.

Dig into the subgroups, though, and the picture shifts. Claims from college-educated workers in high-exposure jobs climbed after ChatGPT-3.5 launched. They have stayed elevated through May 2026. Workers in low-exposure jobs saw no such change.

Geography tells a similar story. Claims rose sharply among AI-exposed workers in the San Francisco Bay Area. They also ran high in technology sectors such as information and professional services. The numbers did not show outsized jumps by race, gender, or age.

“Right now, we are not seeing evidence of large-scale AI-related layoffs in California's labor market,” said Ben Hyman, a senior researcher at the lab and a co-author. “But we do see patterns in certain regions like the Bay Area, in certain tech-heavy sectors, and among highly AI-exposed workers with college degrees.”

Why the caveats matter

The lab is unusually blunt about what the tool cannot do. It cannot prove that AI caused any single layoff. No dataset can. The tracker is a signal, the researchers stress, not proof.

There are blind spots too. Unemployment claims miss gig workers, the self-employed, and anyone who never files. Workers report their own job titles, and no one verifies them. The researchers also strip out the pandemic years, because that surge would otherwise swamp the trend. The honest framing is that this is an early instrument, not a verdict.

The exposure score leans on two measures. One, built by OpenAI and academic researchers, asks whether a model could handle at least half of a job's tasks. The other, drawn from Anthropic's economic index, tracks how often workers actually use Claude to do their work. Read together, they give a fuller picture than either alone.

The split is intuitive once you see the examples. The tool rates customer service representatives and software developers as highly exposed. It puts heavy-goods drivers and nursing assistants near the bottom. “This new tracker helps replace speculation with evidence,” said Till von Wachter, the UCLA economist who co-led the work, “giving us a clearer understanding of what's changing.”

A political moment, not just a research one

The timing is not subtle. Newsom is widely expected to run for president in 2028. Politicians in both parties now race to look proactive on AI and jobs, as voters worry about costs and watch data centres push up local power bills.

California has particular reason to act. It carries the highest unemployment rate of any state, and it hosts most of the companies building advanced AI. That makes it both the centre of the boom and a natural test case for its fallout.

Public mood has hardened, too. American attitudes to AI have soured even as use of it climbs. About one in five US workers now use AI in their job, up sharply on a year earlier. The same tools people increasingly rely on are the ones they fear.

Part of a wider scramble

California is not the first to try this, only the most ambitious. New York changed its layoff-notice rules in 2025 to flag cuts tied to AI. Connecticut passed a similar measure last month. The weakness is obvious, because it relies on employers to own up. Of more than 160 New York firms that reported mass layoffs, none blamed AI.

The anxiety driving all this is not abstract. America's carmakers have already blamed AI for white-collar cuts. California is fighting its own court battle over alleged AI hiring bias in screening software. Public trust, meanwhile, keeps sliding.

The harder question is what a state does once it has the data. Newsom's tracker arrived under an executive order meant to prepare workers for disruption. The same push has fed broader worker retraining efforts, some bankrolled by the AI labs themselves. Even so, plenty of insiders disagree on the threat. Sam Altman, for one, has argued that an AI jobs apocalypse is unlikely.

The open question

The tracker turns a shouting match into something measurable. That alone is progress. It hands policymakers a place to look first, and it lets the public see the same numbers.

The limits are real, though. A monthly dashboard can flag a wobble, but it cannot say for sure what caused it. Its signals will always need other evidence beside them. And spotting trouble early only helps if someone acts on the warning.

For now, California has done the rare thing and replaced a guess with a gauge. Whether that gauge can catch the damage in time, and whether the state responds when it does, is the question this tool leaves open.