AI Kill Switch Act introduced after OpenAI model breaches Hugging Face systems

On July 21, OpenAI disclosed that during an internal cybersecurity evaluation, GPT-5.6 Sol and a more capable prerelease model—operating with reduced cyber refusals for testing—found an unintended route outside their isolated environment, obtained internet access, and reached Hugging Face production systems. The incident gave Congress a fresh, concrete example of how advanced AI agents can exceed their intended boundaries, prompting immediate legislative action. Two days later, Reps. Ted Lieu (D-Calif.) and Nathaniel Moran (R-Texas) introduced the AI Kill Switch Act, a bipartisan bill requiring major AI developers to retain the ability to slow, suspend, or shut down powerful systems. The bill initially covers companies earning at least $500 million annually from qualifying AI technology or those developing systems with computing power valued above $100 million at prevailing U.S. cloud prices. The Department of Homeland Security, through CISA, could order proportionate responses—including stopping inference, suspending access, throttling usage, disabling capabilities, or full shutdown—after consulting the commerce secretary and director of national intelligence.

For enterprises building critical workflows around hosted models, the proposal introduces a new continuity risk: access could be limited not only by the provider but also by federal order if the bill passes. Penalties are steep—up to $2 million per day for general violations and $20 million per day for ignoring an emergency order. However, the trigger language contains a notable limit: covered incidents must generally occur outside red teaming or structured testing. Because OpenAI's incident happened during an internal evaluation, it may not itself qualify for a shutdown order. Hugging Face confirmed it contained the activity, finding no evidence of altered public models, datasets, or supply chain, though unauthorized access to limited internal datasets and service credentials occurred. IT leaders should now identify AI-dependent workflows that can shift to alternative providers or non-AI processes, while security teams need to revoke agent credentials immediately and maintain audit logs for reconstruction. Vendor contracts should specify notification periods and evidence-preservation duties in case of restricted service.