Newsom signs California SB 947 to block AI-only firing and

California Gov. Gavin Newsom signed SB 947 on Sept. 30, restricting employers from using automated decision systems as the sole or principal basis for firing or disciplining workers. The law adds a human-corroboration requirement and employee-facing disclosures, but leaves an undefined “primarily relies” trigger that business groups say creates compliance uncertainty.

Key Takeaways

California enacted SB 947, the “No Robo Bosses Act,” barring employers from relying solely on automated decision systems to fire or discipline workers and limiting AI’s use as a “principal tool” in those decisions.
When termination or discipline decisions are primarily driven by AI output, the law requires a human reviewer to corroborate the outcome using other information, including managerial evaluations, peer reviews, and personnel files.
Workers affected by an AI-primary decision must receive written notice that AI was primarily used, a description of the employee data used by the system, and a human point of contact who can explain the decision.
Business opposition centered on the bill’s undefined compliance trigger, arguing “primarily relies” lacks an objective standard and could widen disputes over whether a tool was advisory or determinative.

California’s SB 947 Puts a Human Gate on AI-Driven Firings

California Gov. Gavin Newsom signed SB 947, the “No Robo Bosses Act,” on Sept. 30, creating a new set of constraints on how employers can use automated decision-making systems (ADS), meaning software that uses algorithms or AI to help make or automate decisions, in discipline and termination.

Mechanically, the law does two things at once. It bans employers from relying solely on AI or ADS to fire or discipline workers, and it restricts AI’s use as a “principal tool” in those decisions.

The operative compliance hook is “primarily.” If an employer relies “primarily” on AI output to make a termination or disciplinary decision, SB 947 requires a human reviewer to corroborate the decision using additional information, including managerial evaluations, peer reviews, and personnel files. That is a human-in-the-loop mandate aimed specifically at adverse employment actions.

SB 947 also forces disclosure at the point of impact. Affected employees must receive written notice that AI was “primarily used” in the termination or discipline decision, a description of the employee data used by the system, and a human point of contact who can further explain the decision.

Supporters framed the law as a hard stop on machine-only management. State Sen. Jerry McNerney, the bill’s author, said: “No worker should ever be fired or disciplined by a machine, AI or not. Artificial intelligence systems have the potential to boost productivity, but they’ve also made errors and misjudgments and exhibited bias,” adding: “AI must remain a tool controlled by humans, not the other way around.”

The Compliance Trigger Traders Will Fixate On: “Primarily Relies”

The bill’s practical bite is not the headline prohibition on AI-only decisions. It is the interpretive boundary around when an employer “primarily relies” on an automated decision system, because that is what triggers the human corroboration workflow and the written disclosures.

Business groups argue that boundary is not defined in the statute’s trigger language, which turns compliance into a facts-and-circumstances fight. Robert Singleton, the Chamber of Progress’ senior director of policy and public affairs for California and US West, wrote in a letter urging a veto: “The bill’s obligations generally apply when an employer ‘primarily relies’ on an automated decision system, but that critical term is never defined. Employers are given no objective standard for determining when a technology has moved from merely informing a decision to being a primary basis for it,” adding: “Uncertainty about whether ordinary tools qualify as regulated automated decision systems could discourage employers from using technologies that improve consistency, identify safety risks, or help managers make better-informed decisions,”

For traders, that ambiguity is where the near-term risk sits. A rule that is clear on paper but fuzzy at the trigger tends to show up as process cost, audit work, and litigation posture, not as an immediate shutdown of tools. Employers and vendors that sell algorithmic management systems, meaning software used to instruct, monitor, evaluate, or manage workers, may need to build cleaner internal attribution trails that can answer two questions after the fact: what data the system used, and whether the AI output was merely advisory or functionally determinative.

The disclosure requirement makes that second question harder to dodge. If a company has to provide written notice that AI was “primarily used” and describe the employee data used by the system, it needs a defensible record of what the system did and how the human reviewer corroborated the decision with non-AI inputs.

How the Bill Changed After Newsom’s Veto—and Why It Passed This Time

SB 947’s path matters because it explains why California landed on a narrower, operational rule rather than a broad notification regime. Newsom vetoed an earlier version last October, even after it cleared both state legislative chambers with “overwhelming majority support.”

In his veto message, Newsom wrote: “I share the author’s concern that in certain cases unregulated use of ADS [automated decision-making software] by employers can be harmful to workers,” but added that the bill, “rather than addressing the specific ways employers misuse this technology, the bill imposes unfocused notification requirements on any business using even the most innocuous tools.”

McNerney reintroduced the bill in February with changes designed to meet that objection. The revised version removed a pre-notification requirement that would have compelled businesses to alert workers in advance whenever an AI system was in use that could affect work conditions. It also stripped language that would have extended protections to gig workers, a provision that had drawn heavy criticism from rideshare companies including Uber and Lyft.

The revised bill still drew business opposition, but the narrower shape appears to have cleared the governor’s stated concern about always-on notification. Labor groups pushed the signing as a workplace power shift. Lorena Gonzalez, president of the California Federation of Labor Unions, AFL-CIO, said: “When working people organize, we get results. Workers across California have demanded that our state lead the way in regulating AI in our workplaces. And today, we see that begin to happen,” adding: “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.”

The bill also picked up a notable political signal from the other side of the aisle. Steve Hilton, California’s Republican nominee for governor, publicly supported SB 947 and argued it “doesn’t go far enough,” writing: “The California Chamber…says rules requiring human review and basic accountability are too burdensome for employers. What a ridiculous thing for them to say,” and adding: “Employers should not be allowed to use AI to decide whether someone is fired, demoted, loses regular hours, or is shut out of the program they rely on for income.”

Signals to Monitor as California Sets a Template for Other States

The immediate unknown is operational: the provided excerpt does not specify SB 947’s effective date, enforcement mechanism, or penalty structure. Those details will determine whether compliance is a near-term scramble or a slower policy rollout.

The second signal is interpretive. Any California agency guidance, rulemaking, or clarifying language that narrows what “primarily relies” means in practice would reduce uncertainty for employers and vendors, and it would also shape how plaintiffs’ lawyers frame disputes over whether a tool crossed the line from recommendation to decision.

The third is legislative spillover. Similar proposals have been floated in New York, Louisiana, and New Jersey, and California’s signing gives other states a working template built around human corroboration plus post-decision disclosure rather than broad pre-notification.

The fourth is corporate behavior. Large employers that use algorithmic management tools will have to update termination and discipline workflows to document human corroboration and to generate employee notices that describe the data used by the system, which is the kind of change that tends to surface first in internal policy updates and vendor procurement requirements.

My Read: This Is a Governance Tax on Workplace AI, Not a Ban on It

The threshold that matters is not whether companies use AI in HR. It is whether their process can be credibly described as “primarily” driven by an automated output, because that is what forces human corroboration, documentation, and employee-facing disclosure.

If California or an enforcing agency clarifies “primarily relies” into something auditable, the setup starts to look like a durable compliance layer that vendors can productize around. If the term stays undefined, the risk shifts toward disputes over classification and after-the-fact narratives about what role the system really played, which is where process cost turns into litigation cost.

Newsom signs California SB 947 to block AI-only firing and discipline decisions