California Assembly Bill 412, titled in the Legislative Counsel's Digest as "Generative artificial intelligence: training data: copyrighted materials," would require developers of generative AI models to answer individualised inquiries from copyright owners about whether their works were used in training. Introduced by Assemblymember Rebecca Bauer-Kahan on 4 February 2025 as the AI Copyright Transparency Act, with Assemblymember Ash Kalra as co-author, it passed the Assembly in May 2025 and, as of 3 August 2026, sits on the Senate Appropriations suspense file (Source: calmatters.digitaldemocracy.org).
Status and legislative history
The bill was introduced on 4 February 2025 (Source: bauer-kahan.asmdc.org). It was heard in the Assembly Standing Committee on Privacy and Consumer Protection on 18 March 2025 and in the Assembly Standing Committee on Judiciary on 6 May 2025, and reached the Assembly Floor on 12 May 2025. In the second chamber it was heard in the Senate Standing Committee on Privacy, Digital Technologies, and Consumer Protection on 8 June 2026 and in the Senate Standing Committee on Judiciary on 23 June 2026. The operative version is the Senate amendment of 30 June 2026. On 3 August 2026 the bill was referred to the Senate Appropriations suspense file (Source: calmatters.digitaldemocracy.org).
California lawmakers took final public comments during the week ending 7 August 2026 on roughly 30 AI bills ahead of suspense-file votes scheduled for Thursday, 13 August 2026 in both the Assembly and the Senate; AB 412 is among the bills awaiting those votes, alongside SB 813, AB 1988 (the Preventing AI User Self Endangerment Act), SB 867 on companion chatbots in toys, SB 1111 on AI and digital replicas, SB 903 on AI use in mental health therapy, AB 2545 on a worker impact data assessment project, and AB 2713 adjusting the existing California AI Transparency Act (Source: transparencycoalition.ai).
Relation to existing law
The bill would build on AB 2013, the California Training Data Transparency Act, signed in September 2024, which requires developers to post high-level summary documentation about the sources of their training datasets before making a generative AI system or a substantial modification available to Californians (Source: crowell.com; see California AB 2013 — Generative AI Training Data Transparency). Where AB 2013 requires a public, dataset-level disclosure made once per release, AB 412 would add a private, work-level disclosure made on request by an identified rights owner.
Key provisions
As amended in the Senate on 30 June 2026, the bill would (Source: calmatters.digitaldemocracy.org):
| Obligation | Requirement |
|---|---|
| Request mechanism | A developer of a generative AI model must make available on its website a mechanism allowing a rights owner to request information about the developer's use of that owner's covered materials, and to supply registration, preregistration or index numbers for one or more covered materials. |
| Recordkeeping | The developer must document and retain requests received from rights owners for a specified period. |
| Response | Subject to specified exceptions, within 30 days of receiving a request the developer must assess whether it used the requester's covered materials to develop the model and provide the requester with a list of covered materials. |
| Continuing violation | Each day after the 30-day period on which the developer fails to provide the information constitutes a discrete violation. |
| Private right of action | A rights owner who complies with the requirements for submitting a request and is not provided the information may bring a civil action against the developer for specified relief, subject to a notice and cure opportunity. |
The daily-accrual structure and the notice-and-cure step were both present in the operative version; the notice-and-cure opportunity appears as an amendment to the earlier text (Source: legiscan.com).
Exemptions and carve-outs
The requirements would not apply to a model meeting certain criteria, among them a model trained exclusively using data the developer makes publicly available at no cost to users. The bill also provides that it imposes no liability on a telecommunications service, information service, or cable service provider (Source: calmatters.digitaldemocracy.org).
Positions
Bauer-Kahan's introduction framed the measure as increasing transparency around the use of copyrighted materials to train generative systems "ensuring copyright owners can protect their property as California's technology landscape continues to evolve" (Source: bauer-kahan.asmdc.org). Practitioner commentary has characterised the bill as extending AB 2013's summary-level disclosure into per-work copyright disclosure (Source: crowell.com). No industry-association position on this specific bill has been located.
The disclosure the bill would compel is the same category of information contested in the training-data discovery disputes surveyed at AI Copyright Litigation — Analysis: whether a given work is in a given training corpus. A state statutory route to that information would operate independently of the federal fair-use questions discussed at AI Copyright.
Provenance note: provisions here are summarised from the Legislative Counsel's Digest as reproduced by CalMatters Digital Democracy and LegiScan. Citations should upgrade to the enrolled or chaptered text on leginfo.legislature.ca.gov once the bill's disposition is settled and the primary text is ingested.
Open questions
- Whether AB 412 clears the Senate Appropriations suspense file on 13 August 2026 was unresolved as of 9 August 2026.
- The length of the retention period for rights-owner requests is given as "a specified time period" in the Legislative Counsel's Digest, which does not state the figure.