California bars employers from firing workers on AI's say-so alone

A California employer can no longer read your face, your voice, or the cadence of your typing to guess whether you are happy, frustrated, or about to quit, and then act on that guess.

Governor Gavin Newsom on Wednesday signed a package of AI-related bills that forecloses biometric emotion-prediction AI at work, bars employers from firing someone on the say-so of an algorithm alone, and forces a written notice when automated systems drive a mass layoff. The federal floor on workplace AI is empty, and California's package is the most visible attempt by any state to lay one on top of it.

The most visual of the protections is the biometric ban. SB 947, which takes effect July 1, 2027, bars employers from deploying AI that infers an employee's emotions or internal state from face, voice, or keystroke data, the categories that warehouse cameras, call-center audio analytics, and typing-cadence monitors already touch. The enrolled legislative digest also gives workers the right to see descriptions of the data their employer primarily used in any disciplinary or termination decision, and routes enforcement to the Labor Commissioner and public prosecutors. Specified exceptions cover routine administrative use.

The second provision, in SB 951, is narrower than a blanket "AI firing ban." Employers cannot rely solely on AI to fire or discipline someone. AI-assisted decisions, where a human reviews the algorithm's output, are not foreclosed. The law's core is disclosure: when a "covered event," meaning a layoff, termination, or other change driven "wholly or substantially" by AI or other automated technology that replaces or automates employment positions, hits a worker at a covered establishment, the employer must put it in writing. The notice has to name the headcount, the occupation and classification, the location, the job functions the system automated, and the technology category. The state Employment Development Department will publish summaries of those notices and a quarterly statewide tally of reported technology displacements.

That is a real worker protection. It is not a census of AI-driven job loss. The "wholly or substantially" causation standard, the establishment-size threshold, and the explicit inclusion of "other automated technology" mean the eventual numbers will mix generative AI rollouts with the older robotics, scheduling, and forecasting systems that were already in workplaces. The quarterly tally is a starting point for measurement, not a verdict.

California is moving because Washington is not. The Trump administration this week announced a voluntary self-policing accord with major AI labs, and Newsom used the signing to call it out and ask California lawmakers to return for a possible special session on AI. The two tracks are not equivalent. A voluntary accord is a press release. A state statute with a Labor Commissioner, civil penalties, and a published quarterly tally is a system. California is building the second because the first has been the only thing on the table at the federal level.

The new rules do not prevent layoffs, and they do not stop employers from using AI to flag workers for review. They draw a line: emotion inference is out, sole-algorithm termination is out, and a paper trail attaches to every AI-driven mass change. Enforcement will turn on what "solely" and "wholly or substantially" mean in court, and on whether the federal government eventually preempts either statute. SB 951's effective date was not in the legislative text the state has published; SB 947 takes effect July 1, 2027. For now, the most populous U.S. state has answered a question the rest of the country has been leaving open: when an algorithm is part of the decision, what does the worker get to know.