EDD AI and the Labor Market (web)
Citation
California Employment Development Department (EDD). “AI and the Labor Market.” Web page with the California AI-Unemployment Tracker, latest update for September 2026. https://edd.ca.gov/aitracker. Read 2026-11-09. Copy: raw/web/2026-11-09 EDD AI and the Labor Market.html.
Bottom line
- The EDD page hosts the California AI-Unemployment Tracker (CAIT). CAIT counts Unemployment Insurance (UI) claims by how exposed to AI the claimant’s last job was.
- It is a partnership with the California Policy Lab (CPL) and updates each month. The page says the tracker is “a descriptive signal, not causal evidence”.
Background
The EDD runs the state’s UI program. The page links two EDD research briefs on AI and the workforce. It then gives the tracker, its release schedule, monthly updates from July to September 2026, definitions, method, data sources and limits.
Key claims
- “The California AI-Unemployment Tracker (‘CAIT’ or ‘The Tracker’) shown below provides early signs of possible AI-induced job loss.” (California AI-Unemployment Tracker)
- “The tracker was created through a partnership with the California Policy Lab (CPL), a nonpartisan research center at the University of California.” (California AI-Unemployment Tracker)
- “Potential Exposure assesses whether large language models are capable of reducing the time required to complete an occupation’s tasks by at least 50%. Observed AI Exposure captures the extent to which an occupation’s tasks are performed using Anthropic’s Claude platform, based on observed Claude usage.” (Definition of Terms)
- “Using the potential AI exposure measure, the 3-month moving average of high-AI-exposure claims fell by about 2,300 (down about 4.4%), from about 52,200 to 49,900 new initial claims.” (Latest Monthly Update, September 2026)
- “The Tracker should be interpreted as a descriptive signal, not causal evidence, of AI-driven job loss.” (Context for Using These Data)
- “Under this partnership, CPL accesses tabulations from EDD unemployment administrative data in accordance with an agreement governing permitted uses, privacy, and confidentiality requirements.” (About This Research)
Evidence and method
State administrative data: UI claims records from January 2017. Each claimant gets an AI exposure score from the occupation that the claimant reports. The two measures come from Eloundou et al. (2024) in Science and Handa et al. (2025) on arXiv. The pandemic surge from March 2020 to January 2022 is not in the trend figures.
Limits
The page lists its own limits. UI claims leave out workers who do not apply, the self-employed and gig workers. Occupations are self-reported and not verified. The EDD suppresses small cells. The tracker does not measure AI as the cause of a layoff. It does not cover compute or training programs.