This bill adds a new chapter to Title 19 RCW that requires developers of generative artificial intelligence systems (defined in the bill) to post on their websites documentation about the data used to train any covered generative AI system. Developers must post the required training-data documentation on or before January 1, 2027 and again before each time a covered system or a substantial modification to a system (if the system or modification was released on or after January 1, 2022) is made publicly available to Washington residents. The bill creates multiple statutory definitions (for example, developer, generative artificial intelligence, aggregate consumer information, substantial modification, synthetic data generation, and security and integrity) and excludes from the developer definition entities that only develop AI for internal use or research, public entities, tribal nations, and others noted in the text. The required documentation items are specified at a high level and include sources of datasets; how datasets further the system’s intended purpose; counts or ranges for number of data points; high-level types of data points; whether datasets include copyrighted, licensed, public domain, personal information, or aggregate consumer information; whether datasets were purchased or licensed; whether and how datasets were cleaned or modified; dates of first training or last major update; whether synthetic data generation was used; and a detailed description of steps taken to eliminate and remove child sexual abuse material from new and existing training datasets. The posting requirement does not apply to systems whose sole purpose is security and integrity, sole purpose is operation of aircraft in the national airspace, or systems developed for national security/military/defense and made available only to a federal entity; systems regulated under the federal Food, Drug, and Cosmetic Act are also excluded. Compliance is tied to the generally acknowledged state of the art and may include NIST guidance while protecting intellectual property and trade secrets. The bill declares violations to be unfair or deceptive acts or methods of competition under the Washington Consumer Protection Act (chapter 19.86 RCW). The bill references other statutes (RCW 19.373.010 for “personal information,” RCW 19.146.010 for “affiliate,” and federal Food, Drug, and Cosmetic Act provisions) but does not include those definitions. The text provided does not define “foundation model,” does not specify enforcement mechanisms, penalties, specific remedies, or administrative details (such as formatting or website placement), and does not define what it means for a system to be “made publicly available to Washingtonians for use.”

If enacted, companies that develop generative AI for commercial distribution or public use will have to prepare and publish training-data documentation on their websites by January 1, 2027 and again whenever a covered system or a substantial modification (released on or after January 1, 2022) is made available to Washington residents. That means developers will likely need to inventory and summarize dataset sources, general purpose, approximate data counts, types of data, whether data include copyrighted material or personal or aggregate consumer information, whether data were purchased or licensed, data cleaning steps, training dates, whether synthetic data were used, and steps taken to remove child sexual abuse material. The rule excludes certain security, aircraft-operation, federal national security/military systems, systems under FDA regulation, and does not force disclosure of trade secrets, with compliance judged against prevailing technical guidance such as NIST. The most affected parties are private-sector AI developers who sell or otherwise make generative systems available to Washingtonians; they will face ongoing administrative and likely legal costs to prepare, update, and attest to these disclosures and carry some risk of consumer-protection enforcement if they fail to comply or provide misleading information. Public entities, tribal nations, internal-use developers, and simple API users without substantial modification are not covered. Important practical details remain unclear from the text provided, including what exactly counts as “made publicly available to Washingtonians,” how documentation must be formatted or posted, the precise meaning of “foundation model,” and the specific enforcement process and remedies under the Consumer Protection Act.