California SB 947: AI Discipline Requires Human Review

Under California’s chaptered SB 947, starting July 1, 2027, an employer may not rely solely on an automated decision system to discipline or terminate an employee. If the employer primarily relies on the system’s output, it must direct a human to corroborate the decision using data behind that output or other relevant evidence. Output that cannot be corroborated, or that the reviewer finds inaccurate, incomplete or misleading, cannot be used for the decision.

When an employer primarily relies on that output, it must also give the affected employee written notice when it communicates the decision. The employee may request a meaningful, objective description of their own data used by the system. Governor Gavin Newsom signed the measure on September 30, 2026, according to Senator Jerry McNerney’s announcement.

Which systems and decisions are covered?

The law defines an automated decision system, or ADS, by what it does. It covers a computational process derived from machine learning, statistical modeling, data analytics or AI that produces a simplified output, such as a score, classification or recommendation, to assist or replace human discretion and materially affect people. A calculator, database, spam filter or antivirus program is expressly excluded. A vendor’s AI label is therefore less useful than tracing whether its output materially influences an employment decision.

The ban on sole reliance, the corroboration duty, the post-use notice and the employee-data request right expressly concern discipline and termination. The statute also refers to deactivation in a provision about the employer’s burden in certain enforcement proceedings. That reference warrants attention where a platform removes someone’s access to work, but it does not rewrite every operative duty as a general rule for all deactivations.

Other restrictions govern ADS use more broadly. An employer may not use one to violate labor, workplace-safety, employment or civil-rights law; infer an employee’s legally protected status; or predict that a worker will exercise legal rights and take adverse action on that basis. Those prohibitions matter even if the system never recommends firing anyone.

What must the human reviewer investigate?

When ADS output is the primary basis for discipline or termination, a person must do more than accept its recommendation. The reviewer must corroborate the decision with data collected or used to produce the output, or with other relevant supporting information. The law lists supervisory or managerial evaluations, personnel files, employee work product, peer reviews and witness interviews as possible sources; relevant online customer reviews may form part of witness evidence.

The statute calls this duty corroboration rather than prescribing a formal investigation procedure. It allows the reviewer to examine the system’s underlying data, so it does not always demand a separate source of evidence. The decisive question is whether the human can substantiate the decision. If the output cannot be corroborated, or the reviewer finds it inaccurate, incomplete or misleading, the employer must leave that output out of the disciplinary or termination decision.

The statute gives no numerical threshold for “primarily relies.” An employer therefore needs to understand the output’s actual role in the decision, including cases in which a person makes the final selection after a score or recommendation has driven the process. Merely putting a human name on the final decision does not perform the evidentiary check the law describes.

Worker checklist: notice, data and retaliation

The post-use notice is due when the employer tells the employee about a decision made primarily in reliance on an ADS. It must be a separate communication in plain language, in the language used for routine employee communications, and delivered through a simple written method. The law permits email, a hyperlink or another written format.

Look for a statement that the employer primarily relied on an ADS and that a human reviewed the decision and corroborated the output.
Use the human contact named in the notice to seek more information about the decision. The notice must also explain the right to access a description of the employee’s own data used by the ADS.
Request a meaningful, objective description of that data if the system was primarily used for discipline or termination. The employer must anonymize personal information belonging to customers, other employees and other individuals.
Raise a concern or complain to the Labor Commissioner without retaliation. The law also protects attempts to use these rights, allegations of a violation and cooperation with an investigation or prosecution.

The notice must tell the employee that retaliation is prohibited. The protection covers discharge, threats of discharge, demotion, suspension and other discrimination or retaliation. The data right is a right to a description of the employee’s own data used by the system; the statute does not describe it as a right to every underlying record or to the system’s code.

Employer checklist: make the decision reviewable

Preparation starts with identifying where automated output enters performance management and where it becomes a primary basis for discipline or termination. A K&L Gates employment-law analysis recommends identifying those tools and setting procedures for human review, notices and employee-data requests before the rules become operative.

Identify scores, classifications and recommendations used in covered decisions, including output from a contractor’s tool. The statute’s employer definition includes entities that exercise control over wages, hours or working conditions, as well as specified public employers and labor contractors.
Remove any path that makes discipline or termination depend solely on ADS output. Where reliance is primary, assign a reviewer who can examine underlying data or relevant supporting information and withhold output that fails corroboration.
Prepare a separate notice for the moment the decision is communicated. It needs to state the system’s role, human review, a human contact, the data-description right and the prohibition on retaliation.
Set a process for describing the affected employee’s own data while anonymizing other people’s personal information. Preserve a record of what the reviewer examined as a practical way to show how the decision was corroborated.

The notice rule applies after a covered use, at the point the employee learns of the decision. It is not a general advance-notice requirement for every workplace system. Employers also need to assess the separate prohibitions on unlawful uses, protected-status inference and adverse predictions about the exercise of legal rights.

Enforcement and narrow exceptions

The Labor Commissioner may investigate, order appropriate temporary relief, issue citations and bring a civil action. Public prosecutors may also enforce the law. A violation carries a $500 civil penalty per violation. The statute also permits specified court remedies, including injunctive relief, punitive damages and reasonable attorney’s fees and costs in enforcement actions.

In certain proceedings, once use of an ADS for a disciplinary or deactivation decision is shown, the employer must demonstrate either that it did not primarily rely on the system or that it complied with the review and notice provisions. That allocation makes the employer’s account of how the decision was reached consequential if its process is challenged.

A collective bargaining agreement displaces these provisions only if it clearly and explicitly waives them, addresses wages or earnings and working conditions, and protects against algorithmic management. A separate, limited exception covers ADS use required by, or reasonably necessary to comply with, specified federal obligations involving aircraft development or national security, military, space or defense work. That exception reaches only the operations covered by the relevant federal law, regulation or binding contract.