From July 1, 2027, a California employer may not rely solely on an automated decision system when deciding to discipline or terminate an employee. If the employer primarily relies on the system’s output, a person must corroborate the decision using the underlying data or other supporting information, and the employee must get a written notice afterward. The penalty is $500 per violation. The law applies to California employers regardless of size — there is no small-business exemption.
The law is SB 947, the No Robo Bosses Act, approved September 30, 2026. It adds Part 5.5.5, “Automated Decision Systems in the Workplace,” to the California Labor Code. This article walks through what the statute says, what it does not say, and the questions a small employer can settle before July 1, 2027. It is general information, not legal advice. For how the law applies to a specific tool or decision, ask employment counsel.
What the law requires
- No sole reliance. An employer “may not rely solely on an ADS when making a disciplinary or termination decision.” If the system’s output is the whole basis for the decision, that is the violation.
- Human corroboration when primarily relying. If an employer primarily relies on the output of an automated decision system, a human must corroborate the decision using the underlying data or other supporting information. A manager rubber-stamping the output is not corroboration; the person has to look at what the system looked at.
- Written notice after the decision. The employee must receive a separate, plain-language written notice with a human contact. The employee can also request a description of the data the system used.
- The burden is on the employer. If compliance is questioned, it is the employer’s duty to establish it. That means the record matters: what the system produced, what the person checked, and what the notice said.
- $500 per violation. The statute sets a $500 civil penalty per violation. It does not define what counts as a single violation, so do not assume it is per employee or per decision. That question is unanswered in the text.
- No retaliation. An employer may not discharge, demote, suspend, or discriminate against an employee for asserting rights under the law.
Two carve-outs matter for Valley employers. The law does not prohibit automated systems used in developing aircraft for use in the national airspace, or in developing products or services for national security, military, space, or defense purposes, where the use is reasonably necessary to comply with a federal statute, regulation, or binding federal contract. If your shop does defense development work, that carve-out may cover some of your systems — but it does not cover routine employment decisions about your own people. Get the boundary right with counsel before assuming a system is outside the law.
What counts as an automated decision system
An automated decision system, for this law, is a system that scores, flags, ranks, or otherwise evaluates employees in a way that feeds a discipline or termination decision. The familiar examples: productivity scores, attendance flags, quality-monitoring outputs, and driver or picker performance rankings that a manager uses to decide who gets written up or let go.
The law excludes ordinary tools that do not make or feed employment decisions: spam filters, firewalls, identity and access management, calculators, and similar. The question is always what the output is used for, not whether the tool is “AI.” A spreadsheet that ranks employees by a formula and drives terminations is closer to the law’s concern than a chatbot that drafts a memo.
What the law does not do
- It does not ban AI at work. It regulates one use: discipline and termination decisions. Scheduling, drafting, research, and analysis tools are not the target.
- It does not cover hiring. Screening and selection tools are covered by California’s FEHA automated-decision-system regulations, in effect since October 1, 2025 — a separate set of rules with its own anti-discrimination and record-keeping requirements, applying to employers with five or more employees.
- It does not prescribe a policy format. The law requires a human step and a notice. How you document them — the procedure, the template, the training record — is yours to write, and it should match how your shop actually works.
Six questions to settle before July 1, 2027
- Which tools score or flag employees? Inventory every system whose output touches discipline or termination: workforce software, monitoring tools, quality systems, telematics. Include tools adopted without IT or HR approval.
- For each one: does the output assist or replace the manager’s decision? If the manager would make the same call without the output, document that. If the output is the decision, that is the gap.
- What does the human check? Write the corroboration step into the procedure: which underlying data the person reviews, what counts as enough, and where the check is recorded.
- What does the notice say? Prepare the plain-language written notice now, with a named human contact — not a generic inbox.
- Can the vendor give you what you need? If the data behind a score lives in a vendor’s system, confirm the contract gives you the output details and the underlying data you would need to corroborate a decision or answer an employee’s request.
- Do managers know? The people who write up employees need to know the rule before July 2027, not after the first decision under it. A short training record is part of the evidence.
Mistakes we expect to see
- “We’re too small.” Size is not the test. How the system is used is the test.
- “Our vendor handles compliance.” The duty sits with the employer. A vendor can supply data and documentation; it cannot corroborate your decision.
- “The system flagged them” as the whole record. Under SB 947, that sentence is the problem, not the defense. The record needs the human step.
