Get Full Government Meeting Transcripts, Videos, & Alerts Forever!
Get email alerts on the Ai Copyright Transparency topic
No spam. Unsubscribe anytime.
Assembly committee advances bill giving creators a right to learn whether their work trained AI models
Summary
Assemblymember Bauer‑Kahan’s measure to give copyright holders a way to learn whether their works were used to train generative AI models won committee approval and was sent to the Judiciary Committee after the Assembly Privacy and Consumer Protection Committee voted to pass the bill as amended.
Assemblymember Bauer-Kahan’s measure to give copyright holders a way to learn whether their works were used to train generative AI models won committee approval and was sent to the Judiciary Committee after the Assembly Privacy and Consumer Protection Committee voted to pass the bill as amended.
Supporters said AB 412 does not change federal copyright rules but gives creators notice when their registered copyrighted works appear in training datasets. "Everything generated by AI originates from a human creative source," Jolie Fisher, secretary-treasurer of SAG-AFTRA, told the committee, saying creators deserve knowledge about how their work is used.
The bill would create a process for copyright owners to submit a fingerprint — a cryptographic signature or “hash” — of a registered work and require covered model developers to check their training collections and report whether that signature appears. Jay Jesima, co-founder of the Transparency Coalition, described fingerprinting techniques used in other industries and told the committee the approach is technically feasible. "One of the first things you do when you look at a large dataset is remove duplicate copies," Jesima said, adding that fingerprinting and deduplication are longstanding practices.
Opponents, including the Electronic Frontier Foundation and the California Chamber of Commerce, warned the mandate to track and disclose matches against U.S. Copyright Office records could be costly and unworkable, give large companies an advantage, and risk overbroad blocking if fingerprinting is used like content-ID systems. Becca Kramer of the EFF said the bill "imposes an impossible new regulatory regime" that could hurt research and startups.
Committee amendments narrowed the bill’s scope to model developers that train large models (not downstream licensees that embed existing models) and added technical language about fingerprinting and response timelines. The author said amendments were intended to make the requirement feasible and to address concerns from industry and rights holders; she also said discussions would continue.
The committee voted to pass AB 412 as amended to the Judiciary Committee. The roll call on the motion appears in the hearing record. The author said the bill builds on disclosure requirements in last year's bill by Assemblymember Irwin (cited in committee as AB 2013), but takes an additional step to allow individual copyright owners to ask whether a specific registered work was used in training.
The bill’s next stop is the Judiciary Committee; the author and supporters said they will continue technical discussions with industry and rights-holder groups while the bill advances.
Sources and attribution: Assemblymember Bauer-Kahan (author); Jolie Fisher, SAG‑AFTRA (support witness); Jay Jesima, Transparency Coalition (support witness); Becca Kramer, Electronic Frontier Foundation (opposition); Ronak Deilami, California Chamber of Commerce (opposition).
