California Bans AI-Only Firings With the No Robo Bosses Act
California Governor Gavin Newsom signed the No Robo Bosses Act on 30 September, making it the first state law to bar employers from firing or disciplining workers based solely on automated decision systems. When AI plays the primary role, a human reviewer must corroborate the decision with independent information, and affected workers get written notice explaining how the system was used.
On this page
The first state ban on machine-only termination
California Governor Gavin Newsom signed SB 947, the No Robo Bosses Act, on 30 September, the legislature's deadline day, making California the first state to outlaw firing or disciplining a worker on the output of an automated decision system alone. State Senator Jerry McNerney, who first introduced the bill in 2025 and reintroduced it on 2 February after Newsom vetoed the previous version as unfocused, called the signature the result of workers insisting that AI remain a tool rather than a judge.
The veto reversal is the political story here. In October 2025 Newsom rejected the earlier SB 7 as too broad; the 2026 rewrite narrowed its scope, cleared the legislature on 31 August, and won a signature on the second attempt. Similar proposals remain stalled in Congress and most other states, which means California is again writing rules that national employers will likely apply nationwide rather than maintain two HR systems.
What the law actually requires
The act amends California's Labor Code and defines an automated decision system as any computational process derived from machine learning, statistical modeling, data analytics, or artificial intelligence that replaces human discretion in decisions significantly affecting workers. Employers cannot rely solely on such a system for discipline or termination. Where an employer primarily relies on one, a human reviewer must corroborate the conclusion using independent information such as managerial evaluations, peer reviews, or personnel files.
Workers are not left to guess what happened. An affected employee must receive written notice that an automated system was primarily used, a description of the employee data it drew on, and the contact details of a human who can explain the decision. Workers may also request the data their system holds about them, and enforcement runs through the labor commissioner, public prosecutors, and a private right of action with civil penalties per violation.
What changed between the veto and the signature
The 2026 version is deliberately narrower than the bill Newsom rejected. It dropped an advance-notification requirement and removed gig-worker provisions that had drawn opposition from platform companies, trading breadth for the votes that mattered. Business groups still object that the trigger phrase, primarily relies, is never objectively defined, which leaves employers to guess how much AI involvement forces human corroboration.
For a blog read by people who build and run these systems, the design choice worth copying is the corroboration requirement rather than the ban. The law does not say a model cannot score performance or flag policy violations; it says a consequential decision about a person needs a second source of evidence and a named human who can account for it. That is the same pattern safe agent deployments use: automation proposes, a reviewable process disposes, and the audit trail names who decided.