On September 30, 2026, the last day he could act on bills, California Gov. Gavin Newsom signed a package of AI workplace laws. According to the Associated Press, the laws bar employers from using biometric data to predict a worker's emotional state, require written notice to workers when AI is responsible for mass layoffs, and bar employers from relying on AI to decide to fire someone. The centrepiece, SB 947, the "No Robo Bosses Act," reverses Newsom's 2025 veto of a similar bill.
The strongest case for the law
The case for SB 947 is not hypothetical. The bill's author, Sen. Tom McNerney, points to more than 550 "bossware" products on the market and documented wrongful terminations caused by algorithmic errors. A worker fired because a scoring model misread a productivity log has little recourse if no human ever looked at the file. Requiring a person to check the machine's output is a modest, technology-neutral safeguard. It does not ban AI in HR. It asks employers to do what any responsible manager already does.
The bill also passed with bipartisan margins: 53-14 in the Assembly and 28-10 in the Senate, according to the same Senate press release. That is not the profile of a fringe measure.
What the statute actually does
The bill text on the Legislature's site is narrower than the "robo boss" branding suggests. Employers cannot rely solely on an automated decision system (ADS) for discipline or termination. Where an ADS primarily influences such a decision, a human must corroborate its output using sources such as supervisory evaluations, personnel files, work product, peer reviews, or witness interviews. Employers must also give written notice that an ADS was used, that a human corroborated the result, and whom to contact with questions. The law also bars using an ADS to infer a worker's protected status or to predict and penalise the exercise of legal rights.
Enforcement runs through the Labor Commissioner and public prosecutors, with $500 per-violation penalties plus injunctive relief and attorney fees. The Senate author's office says the bill creates no private right of action. The law becomes operative July 1, 2027, which gives employers nine months to prepare.
That design is a real improvement on the more sweeping alternatives circulated nationally. A corroboration standard, with no ban on the tools themselves and no private litigation engine, is closer to proportionate regulation than most AI bills this session.
Where the drafting is weak
The trouble is in the verbs. "Primarily influences" and "corroborate" are undefined in any operational sense. If a manager skims an AI-generated flag and then pulls a personnel file, has the employer corroborated or merely rubber-stamped? The statute's own list of acceptable evidence is broad, which helps employers, but it also means the human check can be a formality. That cuts both ways. Workers may get little real protection, while employers face litigation risk over whether a given review was substantive enough.
There is also a scope question. The definition of an ADS in the legislation reaches technologies that produce a score, classification, or recommendation that assists or replaces human decision-making. That language covers far more than frontier AI. A spreadsheet-based attendance scoring formula could qualify. A small business with a basic scheduling tool may not realise it is regulated until a complaint arrives.
Newsom's own 2025 veto message, as summarised by Covington's Inside Global Tech, cited "overly broad restrictions" and pointed to forthcoming California Privacy Protection Agency regulations on automated decision-making. Whether this year's text cures that breadth, or only adds exemptions for collective bargaining waivers and certain federal defense contracts, is a question courts and the Labor Commissioner will answer after 2027.
The surveillance and layoff provisions
The companion measures are more of a mixed bag. According to the same Covington summary, a Cal/WARN amendment (SB 951) extends mass-layoff notice duties to layoffs caused "in whole or substantial part" by an AI system, with disclosure of affected job classifications and the technology category, operative January 1, 2027. Notice is a light-touch, information-forcing tool, and it fits the evidence-based approach. But attributing a layoff "substantially" to AI is a causal claim employers will struggle to make cleanly, and a company that cites cost pressure instead of AI faces little scrutiny. The rule could reward vague disclosure.
The surveillance bill (AB 1883) bars systems that recognise or infer emotional states and collect neural data, with exemptions for safety and federal compliance. Banning emotion inference rests on a solid scientific objection: the evidence that facial or biometric signals reliably reveal inner states is weak. A prohibition on a technology that does not work well is easy to defend. The risk is in the margins, such as wellness tools that workers opt into or accessibility features, where the boundary between "inferring emotion" and legitimate monitoring is unclear.
The federal backdrop
Newsom used the signing to attack the White House for failing to pass comprehensive AI rules, and AP reports he suggested a special legislative session may follow. The criticism has force. California is filling a vacuum, and a patchwork of state rules is the predictable result. But a patchwork is also a cost for any employer operating across states. If Congress wants to pre-empt, it should do so by passing a substantive standard, not by relying on voluntary industry pledges.
What to watch
The law's value will depend on implementation. Regulators should publish guidance before July 1, 2027, defining what counts as meaningful corroboration, and the Legislature should consider a clarifying amendment narrowing the ADS definition to systems that materially affect employment outcomes. Without that, the human-in-the-loop principle, which is sound, will be tested through enforcement actions instead of clear rules.