Generative artificial intelligence: training data: copyrighted materials.

The deadline for a developer to respond to a rights owner's request for training data information was extended from 7 days to 30 days.

The exemption provision was restructured from applying to a 'developer' that meets criteria to applying to a 'GenAI model' that meets criteria, shifting the unit of analysis from the entity to the specific model.

A new exemption was added for models trained exclusively using datasets that a third party makes publicly available at no cost to users, as disclosed by the developer.

The research exemption was narrowed from any entity developing and using a model 'exclusively for noncommercial academic or governmental research' to only models 'developed and used solely by universities or government entities' for that purpose.

A new exemption was added for models trained exclusively using covered materials that the developer licensed specifically for the disclosed purpose of training a GenAI model.

A new exemption was added for models developed and used exclusively for the operation of aircraft in the national airspace.

A new section (3118) was added providing that nothing in the chapter imposes liability on providers of telecommunications services, information services, or cable services (as defined in 47 U.S.C. Section 153) for content provided by another person, to the extent the provider is not acting as the developer of a GenAI model.

An explicit exclusion was added to the definition of 'covered material' stating it does not include materials that are in the public domain.

The definition of 'rights owner' was simplified from a two-part definition (copyright owner or pre-1972 sound recording owner) to simply 'the owner of a covered material.'

Removed the detailed definition of 'approximate content fingerprint' (or 'fingerprint'), which had specified four technical criteria including distinctness, robustness to minor variations, inability to reconstruct original content, and capability to identify content in a dataset.

Removed Section 3115.5(a), which required developers to proactively document all covered materials they knew were used to train their model, make reasonable efforts to identify other covered materials, and document the rights owner of each documented material.

Removed the requirement for developers to make available on their website information sufficient to enable a person to generate a fingerprint compatible with covered materials used in training, generated according to widely accepted industry standards.

The request mechanism no longer requires rights owners to provide fingerprints along with registration numbers. Rights owners now only need to provide registration, preregistration, or index numbers for their covered materials.

The 30-day assessment requirement was simplified from a fingerprint-based analysis (assessing whether the material represented by a fingerprint is likely present in the dataset) to a broader obligation to assess whether the developer used the rights owner's covered materials to develop the model. The response to the rights owner was also simplified from two separate lists (documented materials and fingerprint-matched materials) to a single list of identified covered materials.

Added a new 30-day notice-and-cure period in Section 3117(b). Before filing a civil action, a rights owner must notify the developer of intent to sue and provide information sufficient for the developer to identify and cure the violation. If the developer cures the violation within 30 days of receiving notice, the violation is deemed cured and no action may proceed.

The bill text was reformatted from the official legislative document style (with line numbers, page headers, and column layout) to a plain web-based display. All substantive provisions - definitions in Section 3115, request mechanism in 3115.5, assessment requirements in 3116, rate limits in 3116.5, civil remedies in 3117, exemptions in 3117.5, and the telecom safe harbor in 3118 - remain unchanged.

Website navigation elements (skip to content, home, accessibility, FAQ, feedback, sitemap, login, search fields, bill information links, version history dropdown, share options) were added as part of the web portal presentation layer.

Developer response deadline for fingerprint assessment requests extended from 7 days to 30 days, giving developers more than four times as long to respond.

Documentation retention period reduced from 10 years after use to 5 years after use.

New Section 3118 limits rights owners to one request per calendar quarter per developer per GenAI model, unless the new request contains material new information not previously available. Also clarifies a single request may cover multiple materials.

New Section 3117(d) allows developers to ignore requests that lack sufficient identity documentation or that violate the rate-limiting provision in Section 3118.

Civil action requirement now requires the rights owner to have complied in good faith with the new rate-limiting rules (Section 3118) before bringing suit, adding a precondition for litigation.

Exemption section (now 3119.5) expanded from a single exemption (all training data publicly available at no cost) to four: (a) trained exclusively on publicly available free data, (b) developed and used exclusively for noncommercial academic or governmental research, (c) not trained using covered materials at all, and (d) trained exclusively on covered materials the developer itself owns.

New provision in Section 3116(b)(2) allows developers to satisfy the fingerprint-generation information obligation by directing rights owners to an external tool that is free to use, nondiscriminatory, and reasonably accessible, rather than hosting the tool themselves.

Definition of 'developer' broadened from 'a person, partnership, or corporation' to 'a business, person, partnership, corporation, or other entity,' potentially capturing more types of organizations.

Throughout the bill, 'copyrighted covered materials' was shortened to 'covered materials' and 'copyright rights owner' was shortened to 'rights owner,' streamlining terminology without changing the underlying definitions.

The developer definition no longer requires that the system be designed 'for commercial use.' Now any person who designs, codes, produces, or substantially modifies a GenAI system and either uses it commercially in California or makes it available to Californians is covered, broadening applicability.

A new Section 3119 provides a blanket exemption: the entire title does not apply to a developer that makes all of the data used to train its GenAI model publicly available at no cost on its website. This is broader than the prior limited exemptions that only excused certain documentation and request-response obligations.

The documentation obligation now only requires documenting copyrighted materials 'that the developer knows' were used to train the model, but adds a new requirement to 'make reasonable efforts to identify and document any other copyrighted materials' that were used. This shifts from a strict liability standard to a knowledge-based standard with a due diligence duty.

Developers must now make available on their website information sufficient for a copyright owner to generate an approximate content fingerprint for any copyrighted material that might have been used in training. This is a new affirmative obligation not present in the prior version.

Copyright owners requesting information must now provide: proof of identity, copyright registration numbers for each work, and an approximate content fingerprint for each work. A data protection provision also requires that the developer's handling of this information be 'reasonably necessary and proportionate.'

A new defined term 'approximate content fingerprint' was added, described as an abstract representation of digital content that is distinct to the content, robust to minor variations, incapable of reconstructing the original, and capable of identifying content in a dataset.

The documentation retention period was clarified to run for as long as the developer uses the system commercially in California or makes it available to Californians (whichever is longer), plus 10 years, rather than simply 'as long as the system is used commercially.'

The term 'copyrighted material' is replaced with 'covered material' throughout, and the definition is expanded from only works 'registered' with the US Copyright Office to works that are 'registered, preregistered, or indexed.'

The term 'copyright owner' is replaced with 'rights owner' throughout. The definition is expanded to include not just copyright owners but also owners of sound recordings fixed before February 15, 1972, enforceable under 17 U.S.C. Sec. 1401 (the Classics Protection and Access Act).

The definition of 'developer' is broadened by adding 'or other entity' after 'person, partnership, or corporation,' potentially capturing unincorporated entities and other organizational forms.

A new provision (Section 3117(a)(2)(A)(ii)) states that a rights owner shall not be required to provide a registration number, preregistration number, index number, or fingerprint in order to receive the list of documented materials. This removes a practical barrier for rights owners seeking basic information.

The fingerprint generation requirement is made more specific: developers must make available information enabling a natural person to generate a fingerprint that is both (1) compatible with any covered materials used to train the model and (2) generated according to widely accepted industry standards.

The request mechanism now allows a 'third party authorized to act on behalf of the rights owner' to submit requests, not just the rights owner directly. Registration numbers are also expanded to include preregistration and index numbers.

Developers must now assess whether copyrighted material represented by a fingerprint provided by a copyright owner is likely to be present in their training dataset. Developers are exempt from assessing fingerprints not generated according to widely accepted industry standards.

The response to a copyright owner request was expanded from providing a single 'comprehensive list' of copyrighted materials used to train the model, to providing two lists: (1) materials the developer documented under Section 3116(a), and (2) materials that fingerprint assessment suggests are likely present in the developer's dataset.

The requirement for developers to notify a copyright owner within 30 days that their copyrighted materials were NOT used to train the model has been removed entirely.

The website mechanism for copyright owner requests must now specifically allow submission of: documentation establishing identity, a physical or electronic signature, and registration numbers and fingerprints for one or more copyrighted materials.

Section 3116(b) now requires that the information made available to enable fingerprint generation be 'compatible with the developer's GenAI system or model,' adding a technical compatibility condition that was not in the prior version.

Throughout the bill, references to 'GenAI system or model' were narrowed to just 'GenAI model,' potentially excluding standalone AI systems that are not models from coverage.

'Approximate content fingerprint' is now also defined as simply 'fingerprint,' creating a shorter term used throughout the bill.

A new definition of 'copyrighted material' is added, limiting it to materials registered with the United States Copyright Office. This narrows the bill's reach from all copyrighted works to only formally registered ones.

The definition of 'developer' now requires a California nexus: the developer must either use the GenAI system commercially in California or make it available to Californians for use. Previously, any commercial use qualified regardless of location.

The bill's trigger changed from applying 'before a GenAI system is made publicly available to Californians' to applying to any developer that makes a GenAI system 'available to Californians for use,' broadening when the obligations attach.

A new exemption is added: developers who make their training data documentation (listing copyrighted materials and copyright owners) publicly available at no cost on their website are exempt from the requirement to provide a website mechanism for copyright owner requests (Section 3116(b)) and from the entire response-to-request obligation in Section 3117.

Section 3116 was restructured from five flat subdivisions (a)-(e) into three subdivisions with sub-paragraphs: (a)(1) and (a)(2) for documenting materials and copyright owners, (b)(1) and (b)(2) for the website mechanism and request documentation, and (c) for retention. The actual requirements remain identical.

Section 3117(b) language was clarified from 'a material not used to train a GenAI system or model' to 'a material whose materials were not used to train a GenAI system or model,' making the reference to the copyright owner's materials more explicit.