Ro Khanna introduces AI safety bill banning recursive technology

xAI: xAI develops advanced AI models aimed at advancing scientific discovery and understanding the universe. Its systems would be subject to the bill's licensing regime, audits, and restrictions on recursive self-improvement. The proposal names xAI as one of the leading companies requiring new federal safeguards. OpenAI: OpenAI develops advanced artificial intelligence models, including frontier systems with potential for broad economic and societal impact. The proposed bill would subject its models to licensing, audits, and safety standards enforced by a new federal agency. Khanna highlighted OpenAI among key labs requiring enhanced oversight amid escalating safety concerns. Anthropic: Anthropic builds large-scale AI models with an emphasis on safety and alignment research. Its frontier models would fall under the bill's requirements for independent auditing, training approvals, and containment protocols. The legislation positions Anthropic as one of the primary companies targeted by the new regulatory framework. Ro Khanna: Ro Khanna is a Democratic U.S. Representative from California's 17th district, representing Silicon Valley and focusing on technology policy and regulation. He is introducing the Human Control Over AI Act, which establishes strict liability, bans on recursive self-improving AI, and a new federal oversight agency. Khanna modeled the legislation on input from AI safety groups rather than industry executives. Google DeepMind: Google DeepMind conducts research and develops cutting-edge AI systems, including those integrated with Google's infrastructure. The bill would require its models to undergo federal licensing, security testing, and human-control standards. Khanna included DeepMind among the frontier labs needing dedicated agency focus. Palisade Research: Palisade Research is an independent nonprofit that studies AI risks and develops evaluation methods for frontier models. Khanna referenced conversations with the group in crafting the legislation to prioritize civilizational safety concerns. It helped influence calls for embedded auditors and strict penalties for disabling controls. Model Evaluation and Threat Research: Model Evaluation and Threat Research is an independent nonprofit focused on evaluating AI models for safety risks and potential threats. Khanna consulted the organization while drafting the legislation to incorporate perspectives from the AI safety community. The group helped shape provisions on auditing and containment standards. Machine Intelligence Research Institute: Machine Intelligence Research Institute is an independent nonprofit dedicated to research on AI alignment and preventing loss of control over advanced systems. Khanna drew on its expertise to inform the bill's emphasis on human oversight and safeguards. The organization contributed to positioning safety researchers as central voices in the debate. Regulation: The bill creates a new federal agency tasked with licensing, auditing, and setting security standards specifically for frontier AI models from leading labs. Legislation: It joins other proposals such as the bipartisan FRONTIER Act that would embed independent auditors in labs and enable government shutdown of high-risk models. International: The legislation directs the administration to pursue export controls and agreements to prevent adversaries like China from developing dangerous AI systems.