SHB 2503
In CommitteeHouse
AI training data
Regulating artificial intelligence training data.
This status may be delayed. See Action History below for the latest updates.
How does a bill become law?
- Introduced: The bill is filed and assigned a number.
- Committee: A subject-matter committee holds hearings, takes public testimony, and decides whether to advance the bill.
- Floor Vote: The full chamber (House or Senate) debates and votes on the bill.
- Opposite Chamber: The bill repeats the committee and floor vote process in the other chamber.
- Governor: The Governor reviews the bill and decides whether to sign or veto it.
- Signed: The bill has been signed into law.
AI Analysis
This bill requires companies that develop or modify generative AI systems (like chatbots or image generators) to publicly disclose details about the data used to train those systems. It aims to increase transparency around how AI systems are built, especially regarding data sources, privacy considerations, and intellectual property. The law takes effect in January 2027 and applies to systems made publicly available in Washington.
- Requires developers of generative artificial intelligence systems to publicly disclose documentation about the data used to train their systems, including sources, types, volume, and whether personal or copyrighted data were used.
- Mandates disclosure before each public release or substantial modification of a generative AI system, starting January 1, 2027.
- Exempts certain AI systems from disclosure requirements, including those used solely for security and integrity, aircraft operations, national defense, or federal use.
- Clarifies that violations of the disclosure requirements are treated as unfair or deceptive practices under Washington’s Consumer Protection Act.
- Defines key terms such as 'aggregate consumer information', 'generative artificial intelligence', 'developer', and 'substantial modification' to guide application of the law.
Who is affected
- Developers of generative artificial intelligence systems — Must publicly disclose documentation about the data used to train their generative AI systems, including data sources, types, volume, and whether personal or copyrighted data were used.
- Businesses and startups using or building AI tools — May benefit from clearer rules around data use and transparency, but developers must avoid using personal or protected data without proper disclosure.
- Washington residents using AI tools — Will have more visibility into how AI systems they use are trained, helping them make informed choices and understand potential privacy or accuracy concerns.
- Public agencies and tribal nations — May face new compliance obligations if they develop or modify AI systems for public use, unless they qualify for exemptions (e.g., internal use only).
Pro/Con Analysis
Stronger case for benefits
Potential Benefits (5)
Enhances consumer autonomy by requiring transparency about data sources, types, and privacy considerations—empowering Washington residents to make informed choices about which AI tools to use and to hold developers accountable for deceptive practices.
Rights & LibertiesPeopleRef: Sec. 2(1)(a)-(k)Makes violations of disclosure requirements actionable under the Consumer Protection Act, giving Washington residents a legal tool to challenge deceptive AI practices—especially valuable for vulnerable populations who rely on accurate, non-manipulative AI tools.
consumer protectionPeopleRef: Sec. 4Requires developers to disclose whether datasets contain personal information or aggregate consumer information, increasing awareness among users about how their data may be used—especially helpful for low-income and non-English-speaking residents who may be less informed about AI data practices.
privacyPeopleRef: Sec. 2(1)(g), (h)Requires disclosure of whether datasets include copyrighted material or were licensed—helping creators, educators, and small businesses understand potential IP risks when using AI tools, and supporting fair use claims.
Rights & LibertiesPeopleRef: Sec. 2(1)(e), (f)Requires high-level summaries of datasets and data types used, which can support AI literacy in schools and community settings—especially beneficial for public schools and community colleges teaching digital literacy or computer science.
EducationLean peopleRef: Sec. 2(1)(c), (d)
Potential Concerns (5)
Mandates disclosure of whether datasets include personal information or aggregate consumer information, but does not prohibit use of such data—only requires disclosure—so consumers may learn about data use but gain no protection against misuse or re-identification.
privacyPeopleRef: Sec. 2(1)(g), (h)Exempts AI systems used for security, aircraft operations, and national defense from disclosure, reducing transparency for systems that may be deployed in public safety contexts (e.g., predictive policing, airport surveillance), limiting public oversight of potentially high-impact tools.
Public SafetyLean peopleRef: Sec. 2(2)(a)-(c)Requires developers to disclose whether datasets include copyrighted material or were licensed, but does not impose liability for unauthorized use—only disclosure—so copyright holders may benefit more than everyday users, and small developers may face compliance costs without legal clarity.
Business & EmploymentLean peopleRef: Sec. 2(1)(e), (f)Requires compliance based on “generally acknowledged state of the art,” including NIST guidance, but without specifying how to resolve conflicts between guidance and business practices—creating ambiguity for small developers and startups trying to comply.
Business & EmploymentRef: Sec. 3Mandates detailed technical disclosures (e.g., data point counts, processing methods, synthetic data use), which may impose disproportionate compliance burdens on small AI startups and sole proprietors without proportional benefit to end users.
Business & EmploymentRef: Sec. 2(1)(d), (j), (k)
Who Is Most Affected
Large AI developers (e.g., Microsoft, OpenAI, Anthropic) may face modest compliance costs but benefit from clearer regulatory expectations and reduced risk of consumer lawsuits under existing consumer protection frameworks.
Small AI startups and sole proprietors may face disproportionate compliance burdens (e.g., documenting data sources, processing methods), especially if they lack legal or technical staff—though they may benefit from reduced liability risk if they comply.
Washington residents—especially low-income, elderly, or non-English-speaking users—gain greater visibility into how AI systems are trained, helping them avoid tools that misrepresent or mislead, and empowering informed consent.
Public agencies and tribal nations are exempt from the law’s requirements, so they face no new compliance burden—but may benefit indirectly from increased public trust in AI tools developed by private vendors.
Copyright holders (publishers, authors, media companies) gain visibility into whether their works were used in training—though no direct enforcement power is granted, so impact is limited to awareness and potential future legal leverage.