California's No Robo Bosses Act: What SB 947 Actually Requires
An algorithm can flag you. Under a bill sitting on Governor Newsom’s desk, it cannot be the one that fires you.
SB 947, known as the No Robo Bosses Act of 2026, cleared the California Legislature at the end of August and was enrolled on September 4. Newsom has until September 30, 2026 to sign or veto it. If he signs, it would be the first law of its kind in the country, and it changes what an employer has to document the moment an automated tool touches a discipline or termination decision.
What Counts As An Automated Decision System
The bill defines an automated decision system, or ADS, broadly: any computational process built on machine learning, statistical modeling, data analytics, or artificial intelligence that produces a score, classification, or recommendation used to replace or assist human judgment about a worker, where that judgment has a material effect on them.
That definition is wide enough to catch more than obvious cases like an AI performance-scoring tool. Scheduling software that predicts no-shows, an attendance system that auto-flags a pattern, a vendor’s “risk score” fed into a manager’s review queue, all of it can qualify if the output materially shapes what happens to a person’s job. The bill covers independent contractors as well as employees.
The Core Rule: A Human Has To Actually Decide
The central restriction is narrow and specific. Employers cannot rely solely on an ADS to discipline or terminate a worker. Before acting on what the system outputs, a human has to review it and independently corroborate it, meaning check it against something other than the system’s own say-so.
That is a materially different bar than “a manager clicked approve.” Rubber-stamping an ADS recommendation without an independent check would not satisfy the requirement as described in coverage of the bill’s language. The bill also bars two adjacent uses outright: using an ADS to infer a worker’s protected status, and using one to predict a worker’s future behavior in order to take adverse action against them for exercising a legal right.
What Employers Have To Tell Workers
When an ADS is used in a disciplinary or termination decision, the bill requires the employer to give the affected worker written notice after the fact. Workers can also request a copy of their own ADS-related data, capped at once every twelve months.
That written notice is not boilerplate. Coverage of the bill describes an expectation that it identify what the system evaluated and how the outcome was reached, which means whoever drafts it is producing a document that has to survive scrutiny from the worker, the worker’s attorney, and potentially the Labor Commissioner. Getting the facts in that notice exactly right, without vague hedging and without overclaiming what the system actually did, is the kind of writing task where keeping a human in the loop when AI drafts matters as much as the underlying HR decision does.
A Wrivio Context for this notice could say:
Rewrite this as a formal HR notice describing automated-system involvement in a personnel decision. State plainly what the system evaluated and what a human reviewed and confirmed independently. Do not soften, hedge, or add legal conclusions. Keep every date, system name, and factual detail exactly as written.
Press Ctrl+Shift+Space, paste the draft, and check the diff for anything that drifted from what actually happened, since this is a document where precision outweighs polish.
Enforcement, And Where Sources Disagree
Enforcement runs through the California Labor Commissioner, the state Attorney General, or local prosecutors, with civil penalties reported at $500 per violation. Reporting is split on whether the bill also creates a private right of action letting workers sue directly. Some coverage of the enrolled bill describes worker civil claims with the possibility of punitive damages and attorney’s fees; other summaries describe enforcement as limited to the Labor Commissioner, the Attorney General, and local prosecutors, with no private right of action. Until the signed text (or a veto) settles the question, treat that detail as unresolved rather than picking a side.
The Date That Actually Matters Right Now
The obligations in SB 947, if enacted, become operative on July 1, 2027. That is a real runway, and it means nothing here requires an emergency policy rewrite today.
The date to actually watch is September 30, 2026: Newsom’s deadline to sign or veto. He vetoed a different, broader AI safety bill in 2024, so a veto here is not unprecedented, but SB 947 is a narrower, employment-specific bill built around a single principle, human sign-off before someone loses their job, that has drawn less industry pushback than frontier-model safety rules typically do.
This sits alongside a wider pattern of state-by-state AI employment rules that have been arriving all year, and it is worth reading in that context rather than in isolation: whatever happens with SB 947 specifically, the direction across states is toward requiring a documented human decision-maker, not just a human who was theoretically available to object.
Common Questions
Is SB 947 law yet?
Not as of this writing. The Legislature passed it in late August 2026 and Governor Newsom has until September 30, 2026 to sign or veto it.
Does SB 947 ban AI from being involved in firing decisions?
No. It bars an employer from relying solely on an automated decision system for discipline or termination and requires human review and independent corroboration before acting on the system’s output.
When would SB 947 take effect if signed?
The operative date reported for the bill’s obligations is July 1, 2027, giving employers roughly a year to adjust practices and documentation.
Does an AI writing tool used to draft a performance review count as an ADS under SB 947?
A tool that only assists a human in drafting text, without itself producing the score, classification, or recommendation that drives the decision, falls outside the bill’s core definition, though the final decision-maker should still be the human reviewing and signing off on the outcome.
Where can I read the actual bill text?
The enrolled bill text for SB 947 on the California Legislature’s official site is the primary source for its exact language and current status.
Download Wrivio for Windows to draft the notices and reviews this kind of law requires precisely, with the record staying on your machine when you choose Local mode.
Read Next
New AI Disclosure Laws in 2026: A Simple Way to Sort Them
California, New York, and the EU each added AI disclosure duties in 2026. Here is a three-question test to tell which ones actually apply to your work.
GPAI Code of Practice: Why It Matters When Picking an AI Tool
The EU AI Office can now fine general-purpose AI providers. Whether a provider signed the Code of Practice is a due-diligence signal you can actually check.
EU AI Act Text Labeling Rules: Who Article 50 Actually Covers
Article 50 requires labeling AI-generated text in some cases, not all. Here is exactly which published text needs it, and which everyday drafting does not.
Is the Snapdragon X2 Elite Worth It for Local AI Writing?
The 2026 Snapdragon refresh pushes the NPU to 80 TOPS. For running a small local writing model, here is what that buys you and what it does not.
This article is filed underPrivacy & Compliance, which has 85 articles.