Invest1 publisher2 min readPublished
California requires a human check on firings and discipline driven mainly by AI
Gavin Newsom signed SB 947, requiring California employers that rely primarily on AI to fire or discipline workers to have a human corroborate the decision. The duty attaches to each case, so every AI-led firing or reprimand now needs its own human-assembled evidence file.
The Investor · Invest desk
What happened
- Workers disciplined or fired mainly on AI output must get written notice saying so, a description of the employee data the system used, and a human contact who can explain it.
- Newsom vetoed the bill in October after it cleared both chambers, objecting to a rule that businesses alert workers in advance whenever an AI system affecting work conditions was in use.
- McNerney dropped that pre-notification requirement when he reintroduced the bill in February.
- Former Meta employees allege in a July lawsuit that AI-assisted systems picked workers for layoffs, disproportionately hitting those who took medical or family leave; Meta denies it.
- United for Respect failed last summer to win a Walmart shareholder vote requiring management to disclose more about its use of AI with the workforce.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- decision With the advance-notice rule gone, employers can keep deploying monitoring and scoring tools without new notices; the compliance choice moves to how much weight a tool's output gets in a disciplinary file.
- constraint HR software that cannot report which employee data fed an output becomes harder to use for California discipline, because the employer must describe that data to the worker in writing.
- exposure Disciplined or fired workers get a written account of the data used and a named contact, a record they can bring to disputes like the Meta layoff suit.
The statute allows more than its author's line suggests. "No worker should ever be fired or disciplined by a machine, AI or not," McNerney told CNBC [4]. The law bans relying solely on automated systems for those decisions [1]. An employer may still rely primarily on AI output, provided a human reviewer corroborates the decision with other information [2].
The change that got the bill signed answers the governor's own objection. In his veto message Newsom wrote that the earlier bill "imposes unfocused notification requirements on any business using even the most innocuous tools" [8]. The signed version attaches its obligations to the disciplinary or firing decision itself [2][9].
A lot of software is in play. In an OECD survey published late last year, 90% of US managers said their firms had adopted at least one tool to "instruct, monitor or evaluate workers" [10]. The new duties cover only California decisions where such a tool's output is the main input [2]. The CNBC report does not say how "primarily" is defined or what penalties apply.
That word leaves roughly three ways this plays out. Some employers will record AI output as advisory, keep a manager's judgment on top of every file and send few notices. If most do that, the law changes the wording of HR records more than the practice behind them. Others will accept the trigger, send the notices and build the corroboration step into the workflow. The third route runs through the courts, in cases like the Meta layoff suit [11].
I think the second outcome wins at large employers. Calling AI advisory is a claim the employer has to defend case by case. The law also names the evidence a reviewer would draw on: managerial evaluations, peer reviews and personnel files [2]. An employer that already keeps those records loses little by citing them. The counter-case is that the advisory label costs less and, with the trigger undefined, can be defended. If California employers are sending few notices a year from now while their monitoring tools stay in place, the first route won and this view is wrong.
McNerney first introduced the bill in 2025 after a push by organized labor [5]. The law covers a narrower slice of the disclosure United for Respect sought at Walmart, one decision at a time, and it needed no shareholder vote [12]. "Today, California's workers and our unions have changed the national narrative on how Americans can fight back and win against AI taking over our jobs and workplaces," said Lorena Gonzalez, the bill's lead sponsor and president of the California Federation of Labor Unions, AFL-CIO [6].
What to watch
- Any guidance or amendment defining 'primarily', and the penalty for skipping a notice; a loose definition makes the advisory-label route easy.
- Whether HR software vendors selling in California add per-decision reports of the employee data behind each output to their contracts.
- Notice and complaint volumes from California workers over the next year, the clearest sign of whether employers trigger the rule or label around it.