Virginia files nation’s first AI ‘Do Not Train’ data bill as legislature convenes
Jan. 8, 2025 — Virginia Delegate Michelle Lopes Maldonado wasted no time in Richmond this week, filing four artificial intelligence safety and transparency bills on the eve of the General Assembly’s opening on Wednesday morning.
The four measures take up a variety of issues related to artificial intelligence, including AI disclosure requirements, training data transparency, consumer opt-out mechanisms, safety testing, and deepfakes.
Significantly, Del. Maldonado (D-Manassas) filed HB 2250 as the nation’s first AI safety bill to recognize three emerging consumer protection measures: ‘Do Not Train’ data designations, Training Data Verification Requests, and Training Data Deletion Requests.
Emerging mechanisms for data protection:
Click on image to learn about ‘Do Not Train’ data and Training Data Request/Removal Prompts.
emerging transparency and safety mechanisms
At the Transparency Coalition, we’re strongly backing these mechanisms as appropriate and workable measures to protect both personal data and human-created content. (Learn more about them here.)
HB 2250 is one of several bills nationwide aimed at bringing transparency to AI systems by requiring developers to post information about the training data used to create the models that drive an AI system. California adopted America’s first training data transparency measure (AB 2013) late last year. Virginia’s HB 2250, along with Washington State’s proposed HB 1168, would adopt similar requirements to protect the citizens of their respective states.
Del. Maldonado, a leading voice on AI and tech issues in Richmond, also filed HB 2121, the Digital Content Authenticity and Transparency Act. That bill would require AI developers to embed provenance data in AI-generated or AI-modified images, and provide a tool that would allow public users to discover that provenance. Similar mechanisms are already being used by companies like Adobe, but have not yet become a standard adopted industrywide. (Learn more about that here.)
All four bills, in brief
The four AI-related bills filed yesterday include:
HB 2121: Digital Content Authenticity and Transparency Act.
Requires AI developers to apply provenance data to digital content that is generated by the AI system or service; also requires developers to make a provenance application tool and a provenance reader available to the public. Read the full bill here.
HB 2250: Consumer Opt-Out Artificial Intelligence Training Data Act.
Allows consumers to opt out of the use of the consumer’s personal data. This includes the use of a consumer’s data for the purpose of training an AI model. The bill also requires the developer of a generative AI system to disclose on the developer's website information about the generative artificial intelligence data set used to train such system or service. Requires developers to provide clearly designated and publicly available mechanisms for submissions of Training Data Verification Requests and Training Data Deletion Requests, and to refrain from using clearly designated ‘Do Not Train’ data to train AI models. Read the full bill here.
HB 2094: High-Risk Artificial Intelligence Developer and Deployer Act.
Requires developers of high-risk AI systems to disclose clear information regarding the system’s intended use, purpose, known limitations, foreseeable risks of algorithmic discrimination, performance evaluation results, and risk mitigation measures. Read the full bill here.
HB 2124: Synthetic Digital Content Act.
Expands the applicability of provisions related to defamation, slander, and libel to include synthetic (AI-created or AI-modified) digital content. The bill makes it a Class 1 misdemeanor to use any AI-derived content for the purpose of committing any criminal offense involving fraud. The bill also authorizes the individual depicted in the synthetic digital content to bring a civil action against the person who violates such prohibition. The bill directs the Attorney General to convene a work group to study and make recommendations on the current enforcement of laws related to the use of synthetic digital content. Read the full bill here.
General Assembly’s real work begins next week
Virginia’s General Assembly was briefly gaveled into session on Wednesday morning, but members recessed until next Monday, Jan. 13, due to Richmond’s water crisis. Earlier this week a winter storm knocked the capital city’s water system offline, and facilities in the state capitol building aren’t expected to return to normal until the end of this week.