The Pulse
Four Senators Introduce AI Systems Transparency Act
Sens. James Lankford, Chris Coons, Katie Britt and Brian Schatz introduced the AI Systems Transparency Act on September 24, 2026. The proposal would require certain widely used AI models to publish disclosures about safety testing, user dat
AI.info Team ·
A new Senate bill would require certain AI developers to disclose how their models are tested and what safeguards they use, while also shielding details that could expose security practices or help someone build weapons. Sens. James Lankford, Chris Coons, Katie Britt and Brian Schatz introduced the AI Systems Transparency Act on September 24, 2026, proposing Federal Trade Commission oversight of the disclosures. The draft includes both open and closed models, but sets user thresholds that limit which systems it covers.
The proposal aims to give consumers and independent evaluators more information without compelling companies to publish trade secrets or sensitive security details. That boundary runs through the bill: it calls for public reporting on risks and model behavior, while allowing the FTC to withhold specific information that could compromise protections or materially assist the development of certain weapons.
ASTA sets user thresholds for covered models
The bill would cover generative AI models, including open-weight and open-source models, when they are deployed in systems expected to reach at least 1 million unique monthly users in the United States. A lower threshold of 100,000 monthly users applies when minors can create accounts or profiles. Those thresholds also appear in the bill’s definition of covered deployers, which operate covered systems.
For each covered model, the draft calls for two public disclosures: one designed to give consumers a clear account, and a more detailed version intended to support independent research. Required information would include intended and prohibited uses, capabilities and limitations, supported input and output formats, and whether the system can trigger actions outside the model during use.
Disclosures would cover testing, privacy and misuse
The proposed reports reach beyond the standard description of a model. They would cover safeguards for minors and other vulnerable users, privacy practices involving training and user data, cybersecurity measures, red-team testing and evaluations of risks including offensive cyber use and chemical, biological, radiological and nuclear threats. Companies would also describe the criteria used to decide whether to release a model and report benchmark performance in a form researchers can verify.
The bill calls for updates when a model changes substantially and for reporting on common policy violations, how often systems refuse prohibited prompts, and certain attempts to generate child sexual abuse material or non-consensual intimate imagery. The FTC, consulting with the National Institute of Standards and Technology, would set disclosure rules and define when a model update counts as a distinct model. Under the draft, initial regulations and guidance would be due within 150 days of enactment.
FTC enforcement comes with security limits
The bill would treat violations of its disclosure requirements as violations of rules against unfair or deceptive practices under the FTC Act, giving the commission authority to enforce them. The draft also directs the FTC to provide technical assistance, particularly to small and mid-sized platforms. It does not require disclosure of information the commission determines would compromise a company’s cybersecurity or safety protections, or materially assist the development or deployment of chemical, biological, radiological, nuclear or offensive cyber weapons.
That exception would not let a developer conceal whether it conducted a risk evaluation, or withhold a general account of its findings. The bill’s sponsors frame the measure as a way to make companies’ safeguards more legible to families and researchers. “Trust has to be earned, and right now families are being asked to trust AI companies with very little information,” Lankford said in the announcement.
The proposal follows a December 2025 push
The senators’ release says the bill builds on letters they sent in December 2025 to OpenAI, Microsoft, Google, Anthropic, Meta, Luka, Character.AI and xAI, urging greater transparency about model capabilities and risks. The new proposal would turn that request into disclosure requirements backed by federal enforcement. It remains a bill, not a law; the full draft is available in the published legislation.