The veto he reversed

Newsom killed a nearly identical bill, SB 7, last October, calling it overly broad. Nothing in the underlying technology changed since then. What changed was the politics. Gallup found 39% of Americans now say AI does more harm than good, up from 31% in 2025, and Pew found 71% expect AI to cost people jobs. The revised bill also dropped advance-notice requirements and stripped gig-worker protections, which took the edge off pushback from Uber and Lyft. Sen. Jerry McNerney, who authored both versions, put the case plainly: "No worker should ever be fired or disciplined by a machine, AI or not."

What the law actually requires

Employers can keep using AI for routine tasks. The line is crossed when an automated decision system becomes the primary basis for an adverse action. At that point a human reviewer must independently corroborate the outcome, and the affected employee must be told in writing that AI played the primary role, what data it considered, and who to contact about it. Workers can also request a copy of the data the system used once a year. The law goes further than termination paperwork: it bans systems that predict a worker's future behavior, beliefs, or emotions, and blocks systems that infer protected characteristics such as race or gender for employment decisions.

The part labor didn't get

The bill was weakened on its way to the governor's desk. Contractors are no longer covered. California Labor Federation president Lorena Gonzalez said more work remains, particularly on AI in healthcare. The signing came as part of a package: AB 1883 bans AI tools that predict employees' emotional states or collect neural data, and AB 1331 bans workplace surveillance in bathrooms. In a separate move, Newsom signed an executive order requiring state agencies to spell out "artificial intelligence" instead of using "AI," and criticized the lack of federal AI regulation.

The open question

Here is the part worth watching. The law bars relying on AI "as the primary basis" for a firing, but most large employers already route algorithmic scores through a human who clicks approve. Whether a rubber stamp counts as corroboration is the question the courts will decide. Frank Dickson of Dickson Research framed it well: "A machine can recommend, but a human must decide. The hard part is defining what 'decide' means." For the next nine months, the operational question for California employers is what a defensible human review actually looks like: what the reviewer checked beyond the score, and why. The rest of the country gets to watch that definition get written before copying it.

Sources

  1. [1] cio.comRead source
  2. [2] techstrong.aiRead source
  3. [3] finance.biggo.comRead source