A human has to look at the file
California’s legislature has sent Governor Gavin Newsom SB 947, a bill that would stop employers using an automated decision system on its own to discipline or fire a worker. The Assembly passed it 53-14 on 30 August and the Senate 28-10, according to the office of its author, Senator Jerry McNerney.
The mechanism is narrow and specific. Where an employer relies primarily on the output of an automated decision system for a disciplinary or termination decision, the bill requires the employer to direct a human reviewer to conduct an independent investigation and compile corroborating or supporting information for the decision. A rubber stamp is not enough on the face of the text; the reviewer has to go and find something.
Notice, data and penalties
The bill also gives the worker two things after the fact. They must receive written notice at the time of the decision, stating that human review took place and giving contact information for questions. And they may request a description of their own data used by the system in that decision, once every twelve months.
Separately, the bill prohibits using an automated decision system to violate labour law, to infer a worker’s protected status, to run predictive behaviour analysis for employment decisions, or to predict and act against a worker for exercising legal rights. It bars retaliation against workers who use the rights it creates. Enforcement sits with the Labor Commissioner, and civil penalties run to $500 per violation. If signed, the provisions take effect on 1 July 2027.
McNerney’s office framed the bill in one line: AI must remain a tool controlled by humans, not the other way around.
The clock is the story
The legislature adjourned on 31 August having passed around 30 AI-related bills, according to the Transparency Coalition, an advocacy group that tracks the session. Two were signed in June — AB 2148, requiring school employees to be natural persons, and SB 719 on vehicle location tracking disclosure. The rest are on the governor’s desk, and he has until 30 September to sign or veto them.
The others in the pile cover a wide spread. AB 1709 would bar social media companies from offering personalised feeds and autoplay to children under 16. SB 867 would keep companion chatbots out of toys. SB 928 would require California State University instructors to be human. AB 1883 would restrict workplace AI surveillance tools that monitor emotional states, and SB 1119 would update chatbot safety protections.
What to watch
The decisions land in the same four-week window, which makes 30 September the date to mark rather than any single bill. For SB 947 specifically, two things follow a signature: an effective date of 1 July 2027 that gives employers nine months of runway, and a definition of what counts as relying “primarily” on an automated system, which is where compliance arguments will start. A veto would leave workplace automated decision-making in California governed by existing labour and privacy law alone.
