California just started tracking AI job losses

On June 25, California became the first state in the country to put a live number on the question most governments are still avoiding: is AI actually costing people their jobs, and whose? The California AI-Unemployment Tracker is a public dashboard, updated monthly, that cross-references unemployment claims filed with the state against an AI-exposure score for every occupation. When claims climb fastest in the jobs most exposed to AI, the tracker flags it. It was built by the California Policy Lab at UCLA with the Employment Development Department, and ordered into being by Governor Newsom in May. It is the closest thing the United States has to an early-warning system for automation.

Its first reading is worth sitting with. Statewide, there is no evidence yet of large-scale AI layoffs. But claims from college-educated workers in high-exposure Bay Area occupations have risen since ChatGPT arrived in November 2022. The disruption, so far, is narrow, white-collar, and young.

Read the fine print, because the honesty in it matters. The tracker measures correlation, not cause. It can show that claims are rising in exposed occupations; it cannot yet prove that AI, rather than a soft tech market or the long tail of pandemic churn, is the reason. California's own researchers say so plainly: the dashboard flags patterns, not verdicts. That is not a flaw in the tool. It is simply where the measurement stops today, here and everywhere. No government has built an instrument that can pin a single lost job on AI.

This is the gap running through every serious effort. Governments can measure how many firms use AI, and how exposed each occupation looks on paper. What none can measure, in real time, is how much of the job loss AI itself caused. The Federal Reserve Bank of Dallas found a 13% drop in employment since 2022 among workers aged 22-25 in the most exposed occupations, a signal that lines up with California's Bay Area reading. But a signal is not a cause, and a research paper is not a monitoring system.

California is first, but not alone in trying. The rest is a patchwork of the half-built and the stalled. New York passed a bill in June that would make larger employers report AI's effect on hiring and hours; it has sat unsigned on the governor's desk and would not take effect until 2027. Colorado wrote the country's most ambitious AI employment law, then repealed and rewrote it, and enforcement is now frozen by a federal lawsuit. In Congress, a bipartisan bill from Senators Warner and Hawley would require large firms to report AI-related layoffs each quarter; it has not passed, and the current administration has moved to strike down state AI laws rather than write a federal one. Only the Bureau of Labor Statistics has quietly folded AI into its ten-year job projections, which is useful, but a forecast is not the same as a live tracker.

Europe took a different road. The EU AI Act, fully in force from August 2, is the world's most binding law on AI and work, but it governs how employers may use the technology, not whether the technology is destroying jobs across the economy. It puts monitoring duties on companies, not a dashboard in a ministry. Cedefop has surveyed AI skills across member states since 2024; Britain's Department for Science, Innovation and Technology runs an annual labour-market survey. Canada has funded a Statistics Canada program called TechStat but has not yet built it. Singapore and Australia fold the question into their upskilling and safety programs. Everywhere the pattern repeats: plenty of measurement of adoption, almost none of harm.

Now the numbers turn personal. Every honest look at exposure finds women closer to the edge. At the California signing, First Partner Jennifer Siebel Newsom named it plainly: women face a disproportionate risk of displacement as this technology matures. Britain's survey found women's share of AI roles falling from 24% to 20% in five years, thinning out just where the work is growing. Statistics Canada found that jobs more exposed to AI grew 10% for men over the post-ChatGPT period, but only 5% for women.

Here is the twist the headline misses. The earliest hits are not landing on the low-wage, routine jobs the displacement literature has warned about for a decade. They are landing on young, college-educated workers in knowledge roles, the very occupations where women have made their largest gains in twenty years. The ground that took a generation to win is the ground AI is testing first. A warning light is only useful to the people looking at it. Athena members should be the ones looking.

Three cautions, in the house tradition. First, "first" is a narrow claim: California is first with a live state dashboard built on unemployment data, but Europe and Canada have tracked AI's workforce effects in other forms for years. Second, none of these tools, California's included, has solved cause, so treat every displacement figure as a flag, not a finding. Third, the statewide California data showed no significant gender gap yet, which is not the same as no gap; the tracker is built to surface the local, occupational inequities a statewide average hides. The direction, across a dozen studies, is hard to miss. The precision is not there yet.