California AB 2013: AI Training Data Disclosure for US Teams in 2026

What AB 2013 Requires from AI Developers

California's Automated Decision Systems Accountability Act, known as AB 2013, took effect on January 1, 2026, making California the first US state to mandate training data disclosure for AI developers. Under the law, any company that develops or deploys an AI system using data about California residents must publish a summary of the training data used, including data sources, collection methods, and known limitations. The law applies to both commercial AI systems and AI tools used in employment, housing, and lending decisions.

The disclosure requirements are specific. Developers must list the data sources used in training, describe the range of data collected, identify whether the training data contains personal information about California residents, and disclose any third-party data vendors involved in the training pipeline. These summaries must be posted publicly on the developer's website or in documentation that is reasonably accessible to California users. The California Privacy Protection Agency (CPPA) oversees enforcement and can levy fines on non-compliant organizations.

One critical point in the law's initial implementation: disclosures filed in the first months of 2026 showed a pattern of minimal specificity, with developers disclosing data source categories but providing limited detail on annotation vendors, labeling methodologies, or geographic sourcing. Legal analysts expect the CPPA to issue guidance on the adequacy of disclosures in Q1 or Q2 of 2026, which could raise the bar for what counts as a compliant disclosure.

Why the Legal Challenge Failed

Before the law took effect, a coalition of AI industry groups filed for a preliminary injunction to block AB 2013, arguing that the disclosure requirements constituted compelled speech and imposed undue burden on AI developers. A federal judge denied the injunction in December 2025, allowing the law to take effect as scheduled on January 1, 2026. The ruling held that training data disclosure requirements serve a legitimate state interest in transparency and do not violate the First Amendment in the same way as compelled ideological speech.

The failed injunction has practical implications for companies operating in California's AI market. It signals that transparency-based AI regulations, as opposed to outright prohibitions, are likely to survive legal challenges in federal courts. For AI teams building products for California-based users, this means that training data sourcing decisions made today will need to meet disclosure standards that may become more demanding as the law matures and the CPPA issues interpretive guidance.

Compliance Implications for Enterprise AI Teams

Enterprise AI teams in the US face a layered compliance picture as of November 2026. At the federal level, the Commerce Department's AI governance rules, including export control requirements for AI training data involving foreign nationals, add obligations separate from AB 2013. At the state level, California's law is the most detailed, but Colorado, Connecticut, and Texas have all enacted or proposed AI-related transparency measures that could affect training data documentation requirements going forward.

The practical demand from AI teams is for annotation vendors who can provide documentation trails: records of annotator geography, language background, qualifications, any consent mechanisms used, data handling procedures, and quality assurance processes. These records satisfy both AB 2013 disclosure requirements and the broader technical documentation demands of the EU AI Act GPAI provisions, which apply to any AI developer selling into the European market. Maintaining a single documentation standard that covers both regulatory regimes reduces compliance overhead significantly.

One practical gap identified in early AB 2013 disclosures is the treatment of crowdsourced annotation platforms. Developers using gig-economy annotation platforms often lack the documentation depth required to describe annotator selection criteria, language qualifications, or data handling consents in a verifiable way. This is pushing enterprise buyers toward managed annotation vendors who operate documented workforce programs with auditable records.

What This Means for Annotation Program Design

Annotation program design has historically been driven by cost-per-unit and turnaround time. AB 2013 and parallel regulatory developments are adding a third variable: documentability. The ability to describe a training data program in terms that satisfy a legal disclosure requirement is now a purchasing criterion alongside price and speed. This is a structural shift for enterprise AI procurement teams in California and across the US.

The annotation programs that fare best under AB 2013 disclosure requirements are those with clear geographic sourcing records, documented annotator qualifications, defined consent processes for personal data, and an audit trail linking labeled outputs to annotator demographics. Corpshore AI, which manages annotation programs across 35 languages in 80 countries, maintains these records as a standard part of program delivery. All Jwuma-sourced annotation work includes provenance metadata covering annotator location, language, qualification tier, and session-level quality scores, all of which are transferable to the client for disclosure purposes. Contact Corpshore AI at corpshore.ai/start-a-pilot to design a compliant annotation program in under 14 days.

Frequently asked questions

What does California AB 2013 require AI developers to disclose?

AB 2013 requires AI developers to publicly disclose the training data sources used in their AI systems, including data types, collection methods, known limitations, and whether the data includes personal information about California residents. Disclosures must be accessible to California users and are overseen by the California Privacy Protection Agency.

When did California AB 2013 take effect?

California AB 2013 took effect on January 1, 2026. A federal court denied a preliminary injunction that would have blocked the law, so the disclosure requirements are now in full force for AI developers operating in California.

Who does California AB 2013 apply to?

AB 2013 applies to any company that develops or deploys AI systems using data about California residents. This includes companies based outside California if they build AI tools used by California residents or used in employment, housing, or lending decisions affecting California residents.

How does AB 2013 interact with the EU AI Act?

Both laws require training data documentation, but they operate in different regulatory regimes. AB 2013 is a state transparency law focused on disclosure to users. The EU AI Act GPAI provisions require technical documentation for model providers operating in the EU. Building a single documentation standard that covers both reduces compliance overhead for globally active AI teams.

Can crowdsourced annotation platforms meet AB 2013 documentation requirements?

Crowdsourced platforms often struggle to provide the annotator selection, geographic sourcing, and consent documentation that AB 2013 disclosures demand. Managed annotation vendors with documented workforce programs are better positioned to provide the records needed for a complete disclosure, which is one reason enterprise buyers are shifting procurement toward structured vendor relationships.