- jurisdiction
- US-Federal
- level
- federal
- domain
- safety
- status
- proposed
- scope
- Federal bill to prohibit federal agencies from procuring or using artificial intelligence models found to generate child sexual abuse material (CSAM) or non-consensual intimate imagery (NCII), and to prohibit software designed to circumvent AI safety controls. Applies to 'covered applications': AI models the GSA Administrator or a federal agency head determines are non-compliant with NIST AI 600-1 (Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile) obscene/degrading/abusive-content provisions, or that a federal court has determined generated child pornography/CSAM or non-consensual intimate visual depictions of an identifiable adult.
- obligations
- Directs the GSA Administrator and federal agency heads to identify 'covered applications': AI models not compliant with NIST AI 600-1 (the Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile) with respect to non-consensual intimate imagery or child sexual abuse material within 'obscene, degrading, and/or abusive content', or AI models a federal court has determined generated child pornography or non-consensual intimate visual depictions of an identifiable adult. Agencies must remove covered applications from federal information systems and contractor systems within 180 days of enactment (or of a later covered-application determination), and review products and services for covered-application status annually thereafter. NIST must establish performance benchmarks to prevent generation of obscene, degrading, and abusive content, including synthetic child sexual abuse material and non-consensual intimate images of adults, within 90 days of enactment, and a voluntary vendor testing program within 180 days. The bill separately prohibits deliberately manufacturing, importing, or offering technology designed to circumvent technical controls that prevent an AI model from generating child pornography, and creates a private right of action in federal district court for: an individual whose image or likeness appears in child pornography generated by a covered model; an individual injured by a developer's failure to implement sufficiently robust technical controls, or by a person who generated child pornography using their image or likeness; and a developer whose technical controls are circumvented in violation of the Act.
- penalties
- The bill does not create criminal penalties or a fixed civil-penalty schedule for federal procurement violations; the remedy there is removal of the covered application from federal systems, plus vendor certification and congressional-notification duties if an agency implements alternative safeguards instead of removal. For the separate private right of action tied to circumvention technology and unauthorized child-pornography generation, a prevailing plaintiff may elect either actual damages plus the violator's attributable profits, or statutory damages of $50,000; courts may award up to triple damages for repeat violations within 3 years, may grant injunctive relief (with no prior restraint on protected speech), may order impoundment or destruction of the offending device or product, and may in their discretion award attorneys' fees and costs.
- appliesTo
- Federal agencies procuring or using AI models; AI model developers (as defined: any person that develops an AI model, including one who materially modifies and subsequently distributes one)
- notes
- Introduced by Sen. Mark Warner (D-VA) on July 21, 2026 (confirmed via GovInfo's official bill-status record, BILLS-119s5057, S. 5057, 119th Congress), one week before this run's nominal 2026-07-28 discovery-window start, as part of a broader AI legislative package alongside separately-named bills: Data Center Tax Accountability and Disclosure Act, AI AGENT Act (S. 5051), Financial Artificial Intelligence Risk Reduction Act, Secure Artificial Intelligence Development Act (S. 5061), National Workforce Transition Fund, and PHD Talent Act. As of this round, the AI AGENT Act (S. 5051) and the Secure Artificial Intelligence Development Act (S. 5061) have themselves been independently verified against their GovInfo introduced-bill text and are tracked separately in this dataset as us-ai-agent-act-s5051 and us-secure-ai-development-act-s5061; the Data Center Tax Accountability and Disclosure Act, Financial Artificial Intelligence Risk Reduction Act, National Workforce Transition Fund, and PHD Talent Act remain not independently verified and are not included here. Surfacing the SAFE AI Act because it is untracked, real, and in scope: a federal procurement-side prohibition on AI models found to generate CSAM/NCII, distinct from all currently tracked entities, including take-it-down-act (criminalizes publishing NCII/deepfakes and requires platform notice-and-removal, not a federal-procurement ban) and us-no-fakes-act (creates a private property right over voice/likeness digital replicas generally, not CSAM/NCII-specific federal procurement restrictions). Confirmed directly against the full 14-page introduced bill text: prohibition on federal procurement/use of covered applications, NIST performance-benchmark and voluntary testing-program duties, a circumvention-technology ban, and a private right of action with $50,000 statutory damages. CORRECTION (re-verified 2026-09-07 via the official GovInfo bill-status record for S. 5057): the Senate committee referral, previously left blank pending confirmation, is now confirmed. The record states: 'Mr. Warner introduced the following bill; which was read twice and referred to the Committee on Homeland Security and Governmental Affairs,' dated 07/21/2026 (the same introduction date; this only fills in the prior run's 'not yet specified' placeholder). As of this check, no further committee or floor action has been recorded for S. 5057; it remains pending in committee and status stays 'proposed.'
