Back to California

AB-412 • 2026

Generative artificial intelligence: training data: copyrighted materials.

Generative artificial intelligence: training data: copyrighted materials.

Technology
Active

The official status still shows this bill as active or still awaiting another formal step.

Sponsor
Bauer-Kahan
Last action
Official status
Senate - Appropriations
Effective date
Not listed

Plain English Breakdown

The official summary mentions 'specified exceptions' for the 30-day response rule but does not list what they are.

AB-412: Rules for AI Training Data and Copyrighted Works

This law requires developers of generative artificial intelligence to provide a way on their website for copyright owners to ask if their work was used in training the system.

What This Bill Does

  • Requires developers to put a tool on their website so rights owners can request information about how their materials were used.
  • Mandates that developers keep records of all requests they receive from rights owners for a set time period.
  • Orders developers to check if they used specific copyrighted works and send a list back to the owner within 30 days, subject to specified exceptions.
  • States that each day a developer misses the 30-day deadline counts as a separate violation of the law.
  • Allows rights owners who follow request rules but do not get information to file civil lawsuits against developers after being given notice and a chance to fix the issue.

Who It Names or Affects

  • Developers of generative artificial intelligence models made available in California on or after January 1, 2022.
  • Owners of copyrighted materials such as books, music, and other original works.
  • Telecommunications services, information services, and cable service providers are excluded from liability under this bill.

Terms To Know

Generative artificial intelligence
A system or service defined by the law that creates new content based on training data.
Covered materials
Works protected by copyright, noting that sound recordings fixed before February 15, 1972, are not subject to federal copyright but have similar protections under the Classics Protection and Access Act.

Limits and Unknowns

  • The rules do not apply to models trained only on data the developer makes available for free.
  • The text does not list specific dollar amounts or penalties that courts may award in civil actions, only 'specified relief'.
  • Specific exceptions and criteria for when developers must respond are mentioned but not detailed in this summary.

Bill History

  1. California Legislative Information

    Senate - Appropriations

Official Summary Text

Generative artificial intelligence: training data: copyrighted materials.