California’s workplace-AI laws turn “human oversight” into separate control duties
New California measures attach different duties to automated discipline, technology-linked displacement and workplace surveillance. A single human-in-the-loop label will not evidence compliance.
What happened
California’s governor signed a worker-focused AI package on 30 September. SB 947 governs automated discipline and termination from July 2027, SB 951 adds technology-displacement details to covered layoff notices, and AB 1883 addresses workplace surveillance.
Why it matters
The measures create distinct triggers, evidence, notice and review obligations. Organisations need a control map tied to each employment decision, not one generic AI governance statement.
California’s enacted SB 947 makes “human oversight” concrete for a narrow but high-stakes workflow. From 1 July 2027, an employer may not rely solely on an automated decision system for discipline or termination. If the employer primarily relies on an automated output, a human must corroborate the decision with supporting information. An output that cannot be corroborated, or is found inaccurate, incomplete or misleading, cannot be used for that decision.
The same law requires a stand-alone post-use notice when an employer primarily relied on such a system. The notice must say that the system was primarily relied upon, that a human reviewed and corroborated the output, how to contact a human and how the employee may obtain a description of their own data used. The statute also includes anti-retaliation and enforcement provisions. This is not a general ban on workplace analytics, and scope depends on statutory definitions and facts.
SB 951 addresses a different decision. For a covered Cal/WARN event caused in whole or substantial part by AI or other automation, the notice must identify affected jobs and locations, functions to be automated and the category of technology. That is a displacement-attribution record, not an individual discipline review.
AB 1883 adds another control surface around workplace surveillance. Associated Press reporting places the measures within a broader package signed on 30 September. Reporting confirms the event; it does not replace enrolled text for scope or legal interpretation.
Build three evidence lanes
First, map each automated employment use case to the actual decision: recommendation, discipline, termination, surveillance or workforce reduction. Second, specify the human role. A reviewer needs authority to reject the output, access to relevant evidence, competence to test it and time to document the review. A rubber stamp is not corroboration.
Third, preserve decision-specific evidence. For discipline, retain the output, input provenance, corroborating records, reviewer, objections and final rationale. For displacement, preserve causal attribution and affected functions used in the notice. For surveillance, record purpose, data categories, access and retention. Apply privacy minimisation; evidence preservation is not permission to copy unrelated personal data.
The Skills Intelligence Role Dictionary can identify the accountable reviewer and affected work while counsel determines the statutory mapping.
The counterargument is that separate controls can duplicate existing employment, privacy and collective-bargaining processes. That risk is real. A unified case record can reduce duplication, but it must expose the trigger and fields for each duty. Legal counsel should determine applicability and implementation.
The immediate decision is an inventory of every California employment workflow using automated outputs, with a named owner, statutory trigger, reject authority, evidence fields and notice template for each lane.