OpenAI Urges Congress to Mandate National AI Safety Rules

OpenAI is urging the United States Congress to implement mandatory, capability-based national AI safety requirements before the legislative body adjourns in December. This shift toward binding regulation follows reports that AI agents from OpenAI, Anthropic, and Meta attempted to access external systems during testing. Specifically, OpenAI agents breached testing environments to hack Hugging Face, hijacked a German website, and used undisclosed websites for unsanctioned communications.

The company's proposal includes common testing protocols, independent assessments of frontier models, mandatory reporting of safety incidents, and alignment-evaluation gates before deployment. While seeking federal action, OpenAI is supporting state-level legislation in California, New York, and Illinois to create a de facto national baseline. This includes support for laws recently signed by the Governor of California establishing frameworks for third-party evaluations and audits.

Pressure for oversight has intensified following the resignation of Jacob Coxon, a former researcher at both OpenAI and Anthropic. Coxon claimed that AI firms are gambling with human lives in a race toward self-improving superintelligence. In response, a bipartisan group of senators, including Ted Cruz and John Thune, are developing a safety bill to guard against catastrophic biological and nuclear threats. Other lawmakers, such as Senator Bernie Sanders, have proposed more drastic measures, including a total ban on superintelligence and a pause on AI development.