Thune and Klobuchar Prepare Risk-Based AI Safety Bill

U.S. Sens. John Thune (R-S.D.) and Amy Klobuchar (D-Minn.) were preparing a bipartisan AI bill that would classify systems by risk and require disclosure or self-certification, with introduction possible as early as the following week. A Sept. 22, 2023 report described the measure as a response to growing concerns over AI safety and transparency.\n\nThe proposal was still being developed at that stage, but it later became the Artificial Intelligence Research, Innovation, and Accountability Act of 2023, filed as S.3312 on Nov. 15, 2023. The Senate record lists Thune as sponsor and Klobuchar, Roger Wicker, John Hickenlooper, Ben Ray Luján, and Shelley Moore Capito as cosponsors, with the bill referred to the Senate Commerce Committee.\n\nThe framework took a risk-based approach rather than relying on broad government licensing. An earlier draft described by Axios would have assigned the Commerce Department a five-year testing and certification plan. The introduced bill instead required a three-year implementation plan for certifying critical-impact AI systems.\n\nUnder S.3312, covered consumer platforms using generative AI would have to notify users before they interact with AI-generated content. Deployers of high-impact systems would submit design and safety reports before deployment and annually thereafter. Critical-impact organizations would conduct a risk assessment no later than 30 days before making a system publicly available and update it at least biennially. They would also certify compliance with testing, evaluation, validation, and verification standards. The bill text defines critical-impact uses to include biometric identification without consent, direct management of critical and space-based infrastructure, and criminal justice applications that pose significant risks to constitutional rights or safety. High-impact uses include decisions affecting housing, employment, credit, education, health care, and insurance.\n\nThe bill also directs the National Institute of Standards and Technology to research content authenticity and provenance, methods for detecting AI-generated media, and safeguards against anomalous or compromising AI behavior. Its transparency provisions were designed to cover consumer-facing platforms while excluding certain small businesses and nonprofit research platforms.\n\nThe effort emerged as Congress was weighing how to respond to the rapid spread of foundation models following ChatGPT’s public release. Thune presented the approach as a light-touch alternative intended to protect innovation while establishing safeguards for the highest-risk applications. The proposal also reflected a broader Senate debate over whether AI should be governed through licensing, mandatory disclosures, voluntary standards, or sector-specific oversight.\n\nIndustry and policy groups offered both support and criticism. IBM, the American Enterprise Institute, and the Information Technology Industry Council described the framework as a balanced way to preserve innovation while improving consumer protection. The Future of Life Institute later characterized the bill as one of the most comprehensive federal high-risk AI frameworks proposed at the time, but argued that it needed independent external assessments, broader coverage of advanced general-purpose systems, stronger enforcement, and penalties tied more closely to company size. The Mercatus Center warned that relying on a broad and unclear critical-infrastructure framework could create regulatory overreach while still failing to address the most important system-safety risks.\n\nBecause the measure is legislation rather than a deployed product, it has no end-user performance record or benchmark results. Its central practical limitation is structural: companies would remain responsible for much of the assessment and certification process, while the Commerce Secretary could not block public release solely because of a review of a submitted risk report. The Senate record lists the bill as introduced and referred to committee, leaving its legislative progress dependent on further congressional action.