Newsom signs SB 947: human review required before AI-based firings from July 2027
California Governor Gavin Newsom signed SB 947, known as the "No Robo Bosses Act," on September 30, 2026. From July 1, 2027, employers in the state will be barred from relying solely on automated decision systems when making disciplinary or termination decisions — the first law of its kind in the United States. The law requires human investigation and confirmation, gives employees new rights to notice and access, and can be enforced with civil penalties of up to $500 per violation. With nine months until it takes effect, employers, system vendors and insurers face a concrete adaptation deadline — even as lawsuits such as Mobley v. Workday are already testing adjacent AI liability in hiring decisions.
What the law actually requires
The core of SB 947 is a requirement for human participation, not a ban on automation. Where an employer relies primarily on an automated decision system (ADS), a human must conduct an independent investigation and gather corroborating information before the decision is made, according to a review by the law firm Seyfarth Shaw on JDSupra (1fa9ae3d). If the system's output cannot be corroborated, or proves to be inaccurate or misleading, the employer cannot use it as a basis.
In practice, this means an algorithmic flag — for example, low productivity, anomalies in monitoring data, or a system-generated write-up — can no longer be sufficient on its own to discipline or fire someone. The human in the loop must document an investigation of their own that supports the conclusion.
The law also gives employees new rights. According to the Seyfarth analysis, they are entitled to notice after an ADS has been used in a decision affecting them, and to a description of the primary employee data the system was built on (1fa9ae3d).
In addition, the law sets limits on what the systems themselves may do: employers may not use an ADS to infer a worker's protected status under FEHA — the state's equal employment law — or to predict and take action against a worker for exercising lawful rights (1fa9ae3d).
Enforcement and penalties
Enforcement rests with the Labor Commissioner and public prosecutors. Employees can additionally bring private lawsuits, and civil penalties can reach up to $500 per violation, according to Seyfarth (1fa9ae3d). A private right of action combined with a per-violation penalty could create substantial financial exposure in larger layoffs, where each affected decision could in principle count as a separate violation.
The case that left its mark: "Luna"
The bill was authored by Democratic state Senator Jerry McNerney of Pleasanton, who argued that the concern is not hypothetical. According to NewsNation, he cited a case in San Francisco in which an automated system issued a termination notice after machine-based monitoring (688e6bc2).
Time traced the case to Andon Market, an experimental store run by the AI research company Andon Labs. According to BigGo Finance, which builds on Andon Labs' own announcement, the experiment involved an AI manager called "Luna" running store operations — and identifying one employee for termination: a worker who had been late to 17 of 23 shifts. Even so, it was a human who ultimately made the final decision (e6254fe5). None of the experiment's details have been independently verified beyond the company's own account, but the story nonetheless became the legislator's example of how a system can recommend a firing on its own.
The lawsuit casting a shadow over the field
The experiment was not the only backdrop. In July, 26 Meta employees sued the company in the U.S. District Court for the Northern District of California, alleging that AI-based performance evaluations and monitoring data were used to select who would be affected in a layoff of roughly 8,000 people. All of the plaintiffs had, according to the plaintiffs, taken protected leave or requested disability accommodations. Meta has rejected the claims: according to BigGo Finance, the company stated that "the claims are without merit and not based on facts," and that "workforce and organizational decisions were made by humans, not AI" (e6254fe5). The company's statement is unresolved in court; the allegations have not been tested.
The heaviest legal driver may nonetheless lie elsewhere: Mobley v. Workday, a lawsuit concerning AI-assisted hiring, not terminations. According to Insurance Business, a class certification hearing is scheduled for March 2027 — three months before SB 947 takes effect on July 1, 2027. Workday denies any wrongdoing (372d9b6b).
How widespread is algorithmic management?
The figures circulating around the law come from surveys of unclear methodology and should be read with caveats. NewsNation previously covered a Resume Builder survey in which 6 in 10 managers said they had used AI to make decisions about their own staff, including promotions, pay adjustments and terminations (688e6bc2). BigGo Finance also cites an OECD survey from last year said to have found that 90 percent of American companies use at least one algorithmic management tool for task allocation, monitoring or evaluation (e6254fe5). Whatever the uncertainty, both point in the same direction: the systems the law regulates are already in widespread use.
The companion laws against workplace surveillance
SB 947 was signed together with two companion surveillance laws. According to BigGo Finance, AB 1883 bans AI-based emotion inference and the collection of neural data in workplace surveillance, while AB 1331 bans monitoring of employees in restrooms. The surveillance laws are, according to the same source, set to take effect in January — several months before SB 947 (e6254fe5). Taken together, the three laws form a coherent framework: limits on what is monitored, and limits on what can be decided from monitoring data alone.
The insurance market responds
For employment practices liability insurance (EPLI), the law changes the basis for calculating risk. Broker Jonathan Mitchell of Founder Shield points to two problems, according to Insurance Business. First, employers that use external AI tools should treat the liability as their own — vendor liability will not help them at the courthouse. Second is the risk of aggregation: if many insured employers use the same AI platform, a single system failure could trigger losses at many companies simultaneously (372d9b6b).
Vendor contracts are also expected to be reviewed anew. If an employer can no longer rely solely on a vendor's output, the question follows of who is obligated to deliver data documentation, corroboration and access — and who bears the loss if the system proves inaccurate or misleading.
What remains unresolved
Several questions remain open before the July 1, 2027 effective date. The most fundamental is the law's scope itself: Seyfarth describes the law broadly as a ban on relying solely on ADS in "employment-related decisions," while Insurance Business and NewsNation describe it as a ban targeted specifically at disciplinary and termination decisions (1fa9ae3d; 372d9b6b). Whether hiring decisions — as in Mobley v. Workday — are covered in the same way is therefore unclear without the bill text.
In addition, possible exceptions for collective bargaining agreements and federal compliance have been reported, but this currently stands in only one source (e6254fe5). And perhaps most importantly: how "independent investigation" and "corroborating information" are interpreted in practice — how much documentation a human must gather before a decision is legally confirmed — will depend on future enforcement and case law. For employers and vendors, that is likely, not the ban itself, the hard part of compliance.