Skip to content
Dies ist eine Ubersetzung zur Orientierung. Die englische Version ist die massgebliche und rechtsverbindliche Version. Englische Version anzeigen
US-CADEEP coverage

California AB 2013 — Generative AI Training Data Transparency Act: AI Compliance Requirements

California AB 2013 requires any person or company that designs, codes, produces, or substantially modifies a generative AI system made available to California residents (since January 1, 2022) to publicly post a high-level summary of the training datasets used. This applies to businesses that build their own AI models or substantially customize foundation models — not just those using off-the-shelf AI APIs. If your product wraps or fine-tunes a foundation model, you may be a covered "developer." AB 2013 itself specifies no penalty and names no enforcer; non-compliance would most likely be addressed by the California Attorney General under the Unfair Competition Law, and there is no private right of action.

Summary of publicly-available regulatory text as of 2026-08-22. Verify against current official sources before relying on this for compliance decisions. Not legal advice.

Key Facts

Effective Date

January 1, 2026

Maximum Penalty

No statutory penalty — AB 2013 specifies NO monetary penalty amount and names NO enforcer (the statute is silent on enforcement). There is no private right of action. Non-compliance would most likely be pursued by the California Attorney General (or district/city attorneys) under the Unfair Competition Law (Bus. & Prof. Code §17200 et seq.); any monetary exposure derives from that general theory, not from AB 2013 itself. The prior "$5,000 per violation" figure was not in the statute. legal_review_pending.

What Your Business Must Do

3 compliance requirements identified. Critical requirements carry the highest risk of enforcement action.

Training Data Disclosure (AB 2013)

Critical

If you develop, produce, or substantially modify a generative AI system made available to California residents, publish a high-level summary of the training datasets on your website. AB 2013 enumerates the specific items the summary must address: (1) the sources or owners of the datasets; (2) a description of how the datasets further the intended purpose of the AI system or service; (3) the number of data points (ranges are acceptable); (4) a description of the types of data points and how they are labeled; (5) whether the datasets include data protected by copyright, trademark, or patent, or are in the public domain; (6) whether the datasets were purchased or licensed by the developer; (7) whether the datasets include personal information; (8) whether the datasets include aggregate consumer information; (9) whether there was any cleaning, processing, or other modification to the datasets, including the intended purpose of those modifications; (10) the time period during which the data was collected, including whether collection is ongoing; (11) the dates the datasets were first used during development; and (12) whether the generative AI system or service used or continuously uses synthetic data generation. Synthetic-data sets used in that process must themselves be described.

Deadline: January 1, 2026

Cal. Civ. Code § 3111(a) (AB 2013; definitions at § 3110)

Public Website Training Data Summary

High Priority

The training data summary must be publicly accessible on your website — not buried in terms of service. It must be available before or concurrent with making the GenAI system available, and updated when material changes to training data are made.

Deadline: January 1, 2026

Cal. Civ. Code § 3111 ("post on the developer's internet website")

Training Data Summary — Update Before Each Substantial Modification Release

Medium Priority

The update trigger is event-driven, not periodic: Civ. Code § 3111 requires the training-data documentation to be posted "on or before January 1, 2026, and before each time thereafter" that the generative AI system or service — or a substantial modification to it — is made publicly available to Californians. "Substantially modifies" means a new version, new release, or other update that materially changes the system's functionality or performance, including retraining or fine-tuning (§ 3110). So the summary must be current BEFORE a substantially modified version ships, not "promptly after." Maintain version history of disclosures.

Cal. Civ. Code § 3111 ("before each time thereafter"); § 3110 ("substantially modifies")

Who Does This Apply To?

Applies to any person or entity that "designs, codes, produces, or substantially modifies" a generative AI system and makes it available to California residents. Retroactive: covers GenAI systems deployed since January 1, 2022. Key exclusion: companies that merely use or deploy AI via third-party APIs (e.g., calling OpenAI/Anthropic APIs to power a product feature) are NOT covered — only original developers and substantial modifiers. Fine-tuning, RLHF on a foundation model, or retraining on proprietary data likely qualifies as "substantial modification."

Recent Regulatory Guidance

guidance2025-01-13

California AG Legal Advisory — Application of Existing California Law to AI (13 Jan 2025)

On 13 January 2025 the California Attorney General issued a legal advisory ("Application of Existing California Consumer Protection, Civil Rights, Competition, and Data Privacy Laws to Artificial Intelligence") explaining how existing California law applies to AI; it cites AB 2013 (Generative AI Training Data Transparency Act, effective 1 Jan 2026) among the state's AI statutes. The advisory states general principles — it does NOT publish bill-specific "API wrapper vs. substantial modifier" scope carve-outs. To determine whether a product is a covered developer under AB 2013, rely on the statutory text.

Source

Key Case Law & Precedent

X.AI LLC v. Bonta (C.D. Cal., filed 2025-12-29)

US District Court, Central District of California; on appeal to the Ninth Circuit (No. 26-1591) · 2026

The first constitutional challenge to AB 2013 itself, brought by xAI against California AG Bonta: First Amendment (compelled commercial/trade-secret speech), Fifth Amendment Takings Clause, and Fourteenth Amendment Due Process claims. Directly determines whether the § 3111 training-data-summary duty this entry documents is currently, and will remain, enforceable.

Outcome: District Court (Judge Jesus G. Bernal) DENIED xAI's motion for a preliminary injunction on 2026-03-04 — found AB 2013 likely regulates commercial speech (intermediate scrutiny) and xAI had not shown sufficient likelihood of success on the merits, but the order itself flagged xAI retains "a distinct possibility of prevailing on the merits" as the case develops. AB 2013 remains in force; no injunction issued. xAI appealed to the Ninth Circuit (No. 26-1591) on 2026-03-17; oral argument held mid-July 2026; NO RULING had issued as of this cycle (2026-08-22) — pending on appeal.

Case reference

Quarterly Enforcement Digest

Q1 2026: AB 2013 took effect January 1, 2026. The statute is silent on enforcement — it names no agency and specifies no penalty — so it remains unclear which body would enforce it; the most likely path is the California Attorney General (or local prosecutors) under the Unfair Competition Law, with no private right of action. The "API deployers are not developers" reading follows from the statutory definition of "developer" (one who designs, codes, produces, or substantially modifies the system) — it is NOT a published AG confirmation. Primary compliance concern is for GenAI startups building proprietary models or fine-tuning/retraining foundation models on customer data. The required summary must address the 12 statutory disclosure items (sources, purpose, data-point counts and types, copyright/trademark/patent and public-domain content, purchase/license status, personal and aggregate consumer information, cleaning/modification, collection time period, first-use dates, synthetic-data use); format is otherwise flexible and no template is prescribed. Businesses should post summaries now and flag for legal review. Q3 2026 UPDATE (Cycle 21, 2026-08-22): AB 2013 is under a live, unresolved constitutional challenge — X.AI LLC v. Bonta — with a Ninth Circuit appeal (No. 26-1591) pending as of this cycle after the district court denied a preliminary injunction 2026-03-04. The duty remains enforceable today, but developers should track the appeal given the district court's own acknowledgment that xAI retains a real prospect of prevailing on the First Amendment merits.

Digest covers enforcement actions, guidance, and regulatory developments. Last verified: 2026-08-22.

Industry Playbooks covering California AB 2013 — Generative AI Training Data Transparency Act

These industry playbooks include jurisdiction-specific checklist items and guidance for California AB 2013 — Generative AI Training Data Transparency Act.

Frequently Asked Questions

Does California AB 2013 — Generative AI Training Data Transparency Act apply to my business?

California AB 2013 requires any person or company that designs, codes, produces, or substantially modifies a generative AI system made available to California residents (since January 1, 2022) to publicly post a high-level summary of the training… Use Aegis Firma's free scanner to get a personalized assessment in under 5 minutes.

What is the penalty for non-compliance?

The maximum penalty under California AB 2013 — Generative AI Training Data Transparency Act is: No statutory penalty — AB 2013 specifies NO monetary penalty amount and names NO enforcer (the statute is silent on enforcement). There is no private right of action. Non-compliance would most likely be pursued by the California Attorney General (or district/city attorneys) under the Unfair Competition Law (Bus. & Prof. Code §17200 et seq.); any monetary exposure derives from that general theory, not from AB 2013 itself. The prior "$5,000 per violation" figure was not in the statute. legal_review_pending.. Fines are typically scaled by company size, severity of violation, and whether violations were willful or accidental.

How do I comply with California AB 2013 — Generative AI Training Data Transparency Act?

The 3 requirements above cover the core obligations. The fastest path to compliance is: (1) conduct an AI risk assessment, (2) document your AI systems, (3) implement transparency disclosures where required. Aegis Firma generates all required documents automatically.

Official Source

https://leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=CIV&sectionNum=3111.

Last updated: 2026-08-22 — verify at source before relying on this information.

Don't leave compliance to chance

Aegis Firma scans your AI tools, tells you exactly which regulations apply, and generates all required documents — in 30 minutes.

Start your free compliance scan