On September 30, Governor Gavin Newsom signed SB 947, the No Robo Bosses Act. It takes effect July 1, 2027. From that date, California employers may not rely solely on an automated decision system to discipline or fire a worker.
The law defines an automated decision system broadly. It covers any machine learning, statistical, or AI process that produces a score, classification, or recommendation used to assist or replace human judgment. Spam filters, firewalls, calculators, and plain databases are excluded.
Public employers are covered along with private ones. The law names state government, cities, counties, special districts, school and community college districts, and the California State University. The University of California is covered only if its Regents agree.
When an employer primarily relies on a system’s output for a discipline or termination decision, a human has to corroborate that output. The reviewer can use the data behind the output or other evidence: supervisor evaluations, personnel files, work product, peer reviews, or witness interviews. If the output can’t be corroborated, or the reviewer finds it inaccurate, incomplete, or misleading, the employer cannot use it.
The worker gets a separate, plain-language written notice when told of the decision. The notice must say that the employer relied on the system and that a human corroborated its output. It must name a contact person and state that retaliation is prohibited. The worker can also request a description of their own data that the system used.
The law also bars using these systems to infer a worker’s protected status, or to predict and punish workers for exercising their legal rights.
The Labor Commissioner and public prosecutors enforce the law. The civil penalty is $500 per violation. Two exemptions apply. One covers collective bargaining agreements that explicitly waive the law and provide their own protection from algorithmic management. The other covers uses required by federal law or by certain federal defense contracts.
The law does not ban the tools. It bans letting a score stand in for a judgment.
The phrase that matters most is undefined
The review and notice duties apply when an employer “primarily relies” on a system. The law doesn’t define that phrase. Employers will have to draw the line themselves and then defend it. In an enforcement action, once use of a system is shown, the burden falls on the employer. It must prove it did not primarily rely on the system, or that it followed the review and notice rules.
The places to look are wherever a system produces a number that ends up in a discipline file. That includes productivity and activity monitoring, attendance and scheduling flags, call-quality scoring, and driver telematics. Many of these were bought as reporting tools, not decision tools. The law doesn’t ask how a tool was bought. It asks how it was used.
What to do before July 2027
- Inventory every system that produces a score, flag, or recommendation that can feed a discipline or termination decision.
- For each one, decide whether your decisions primarily rely on it, and write down why.
- Define what corroboration means in practice: who reviews, what evidence they check, and how the review is recorded.
- Draft the standalone notice now, in the language you use for routine employee communications.
- Set up the data-access process. Decide who answers requests and how other people’s data gets anonymized.
- Check your vendor contracts. Can the vendor tell you what employee data its system used for a given output?
- Train the people who will do the reviews.
Corroboration is a skill, not a checkbox. Someone has to be trained to do it.
If the decision can’t stand without the score, the score can’t be the reason.
Sources
Pingback: AI Governance Gap: 87% Have It, 22% Say It Works - Brainsquawk