California has become the first state in the nation to launch a public tool tracking whether artificial intelligence is showing up in the unemployment line. The California AI-Unemployment Tracker (CAIT), unveiled June 25, 2026, links unemployment insurance claims data with occupational AI-exposure measures to monitor, in near real time, whether workers in AI-exposed jobs are losing work at elevated rates. The tracker was built through a partnership between the Governor’s Office, the state’s Employment Development Department (EDD), and the nonpartisan California Policy Lab at the University of California, and it is also hosted directly on the EDD’s own website.

The tracker is accompanied by a detailed research report, Tracking AI-Related Job Loss Using Unemployment Insurance Claims Data in California, prepared by a team of researchers at the Policy Lab’s UCLA site, along with a companion technical appendix laying out the underlying methodology. Governor Gavin Newsom framed the launch as part of a broader effort to get ahead of AI’s labor-market effects, saying California intends to lead by “reimagining how we prepare” the state through governance and policy rather than simply watching from the sidelines.

At the center of the tool is a straightforward but data-intensive idea: take every initial unemployment insurance claim filed in California since January 2017, match each claimant’s self-reported prior occupation to a score reflecting how exposed that occupation is to AI, and then track claim volumes over time by exposure level, education, age, gender, race and ethnicity, industry, and region. Claimants are sorted into three exposure tiers — high, moderate, and low — based on the top, middle, and bottom quartiles of the exposure scores. High-exposure occupations, the top 25%, include roles such as customer service representatives and software developers; low-exposure occupations, the bottom 25%, include jobs such as heavy truck drivers and nursing assistants.

The researchers use two separate exposure measures rather than one. The first, called Potential AI Exposure, asks whether large language models are theoretically capable of cutting the time needed to complete an occupation’s tasks by at least half; it comes from a widely cited 2024 study, GPTs Are GPTs, published in Science by a team of OpenAI and academic researchers. The second, called Observed AI Exposure, instead measures how often an occupation’s tasks actually show up in real usage of Anthropic’s Claude, drawing on the Anthropic Economic Index, a 2025 research effort analyzing millions of anonymized Claude conversations. The report notes that results are largely consistent whether the theoretical or the observed measure is used, and cautions that either measure captures only whether a job’s tasks could be or have been touched by AI — not whether AI actually caused any particular layoff. The tracker also excludes the pandemic-era claims surge from March 2020 through January 2022 from its trend comparisons, since that spike would otherwise swamp any post-ChatGPT pattern.

The June 2026 release’s first finding is, on its face, reassuring: looking at all unemployment claims statewide, there is no evidence of a broad-based surge in layoffs among AI-exposed workers since the release of ChatGPT-3.5 in late 2022, and the overall share of claims coming from AI-exposed occupations has not risen in a statistically meaningful way relative to before the pandemic. The report notes this lines up with existing national estimates, including Yale Budget Lab’s analysis of Current Population Survey data, which similarly finds no nationwide link between AI exposure and unemployment so far.

But the tracker’s second and third findings complicate that reassurance considerably. Unemployment claims among college-educated workers in highly AI-exposed occupations rose after ChatGPT-3.5’s release and have stayed elevated through May 2026, even as claims among similarly educated workers in low-exposure jobs showed no comparable shift. The effect is sharpest at the top of the credential ladder: claims among workers with master’s degrees or PhDs in highly AI-exposed occupations climbed from a baseline of about 13,000 per month in November 2022 to a range of 16,000 to 22,000 per month by mid-2023, and have remained in that elevated band since. Geographically and sectorally, the pattern concentrates where AI adoption itself has concentrated: claims from AI-exposed workers in the San Francisco Bay Area show a sharp, sustained increase relative to pre-pandemic levels, and claims in AI-exposed technology sectors — particularly Professional Services — have likewise stayed elevated, though the report notes claims in the Information sector spiked temporarily before settling back to pre-generative-AI levels in late 2025.

Report co-author Ben Hyman, a senior researcher at the Policy Lab, summed up the tension in the findings: the state is not seeing large-scale AI-related layoffs, but is seeing a real pattern among highly educated workers in the Bay Area and in tech-heavy sectors that will bear continued watching. Co-author Till von Wachter, a UCLA economics professor and the Policy Lab’s UCLA faculty director, cast the tool’s value as replacing speculation about AI’s labor-market effects with an evidence base policymakers can act on before disruptions spread further.

The report is careful to flag the limits of what unemployment insurance data can show. UI claims miss workers who never file — because they are unaware of the program, land a new job quickly, leave the labor force, are ineligible because of immigration status or self-employment, or, for younger workers, have not yet accumulated enough qualifying earnings. Occupation codes are self-reported at the time of filing and unverified, small-count data cells are suppressed under standard confidentiality rules, and monthly figures are described as preliminary and subject to revision as late claims are processed. The authors are explicit that the tracker is a descriptive early-warning signal rather than causal proof that AI is driving any specific job loss.

Practically, the tracker itself is a public, interactive dashboard built on Tableau, embedded on the Policy Lab’s site and mirrored on the EDD’s, and it will be refreshed monthly. The underlying tabulated data are also released for public download in an accessibility-compliant Excel file, and the Policy Lab has published a separate FAQ document addressing common questions about definitions and methodology. For California employers and workers’-compensation professionals, the tracker offers an early, if imperfect, gauge of where AI-linked displacement is concentrating by sector and region — useful context for anticipating claims trends even though, as its own authors stress, it cannot yet say whether AI caused any individual worker’s job loss.