California AB 2013: What US Enterprise AI Teams Must Know About Training Data Disclosure in 2026

What California AB 2013 requires

AB 2013 applies to developers of generative AI systems that are offered to California residents. It requires those developers to publish a high-level summary of their training data on their website. The summary must cover the sources used, the types of data included, whether copyrighted material was used, and the approximate date ranges of data collection.

The law does not require disclosure of every individual data source or the precise composition of a training set. It is a transparency measure, not a data registry. However, meeting it in good faith requires that developers have already tracked the information they are disclosing, which means the documentation obligation starts at the data sourcing and annotation stage, not at publication time.

How AB 2013 affects enterprise training data sourcing decisions

The practical effect of AB 2013 is to raise the cost of informal sourcing. If your team has collected training data from public web scrapes, third-party vendors, or contributors without documenting what was collected, when, and under what terms, you now have a disclosure obligation you cannot meet accurately. Retroactively reconstructing that documentation is expensive and sometimes impossible.

Enterprise teams building or fine-tuning models for public use in 2026 and beyond need a sourcing approach that generates documentation as a byproduct of the data collection process, not as a retrospective compliance exercise. This is one of the core advantages of working with a managed annotation partner whose operations are documentation-native.

1. Document data sources at the point of collection, not after the fact.

2. Track copyright status and consent basis for each dataset or contributor class.

3. Record collection timeframes accurately for each data batch.

4. Distinguish between proprietary, licensed, and human-annotated data in your provenance records.

5. Build your disclosure summary before publication, not under deadline pressure.

Related US regulations to monitor

AB 2013 is California-specific, but it is part of a broader US shift toward AI transparency obligations. Colorado's AI Act, in force in 2026, requires developers of high-risk AI systems to provide deployers with documentation including general information about training data categories. At the federal level, there is no comprehensive AI law, but the FTC applies existing Section 5 authority to AI-related deceptive practices, and EEOC guidance covers AI use in employment decisions.

California's SB 942 and SB 853, which extend AI transparency requirements further, take effect January 1, 2027. Enterprise legal teams tracking AB 2013 compliance should include both laws in their 2027 planning cycle, as they build on the data-disclosure framework AB 2013 establishes.

How Corpshore AI builds compliance documentation into managed engagements

Every Corpshore AI managed annotation engagement through the Jwuma platform generates structured provenance records as standard output, not as a premium add-on. Each batch includes the collection date range, task type, contributor language and region profile, and the quality acceptance tier applied. For clients subject to AB 2013, these records map directly to the disclosure categories the law requires.

Clients can export provenance reports for each engagement from the Jwuma client portal. The export is designed to be referenced directly in a training data disclosure summary, reducing the work required to satisfy the AB 2013 documentation standard without additional internal tooling.

1. Provenance records: batch-level documentation of collection dates, task type, and contributor profile.

2. Quality audit logs: acceptance and rejection records for each labeled item, with reviewer tier noted.

3. Consent documentation: records confirming that contributor data is supplied under Jwuma's standard terms, not unlicensed collection.

4. Export-ready reports: formatted for direct use in a training data disclosure summary.

Practical next steps for enterprise AI teams

If your team is building or fine-tuning a generative AI system for public use and you have not yet audited your existing training data for AB 2013 compliance, now is the time to start. The key questions are: what data was used, where did it come from, what was its copyright status, and do you have records covering each of these points.

For ongoing or new data collection, structuring your annotation engagement with a documentation-native partner removes the compliance overhead from your internal team. Corpshore AI can scope a managed annotation programme that generates AB 2013-compatible provenance records from day one.