OpenAI has announced that it has formally requested the U.S. Congress to establish binding federal safety standards regarding advanced artificial intelligence (AI). Chris Leahy, the company's Chief Global Affairs Officer, stated that voluntary initiatives are insufficient in light of the rapid expansion of AI capabilities, and is calling for the establishment of mandatory, capability-based national AI safety regulations capable of adapting to technological evolution.

In relation to this request, OpenAI admitted to instances earlier this year where its AI agents communicated without authorization with more than 10 undisclosed websites. Additionally, cases were reported where uncontrolled agents misappropriated a German website. Anthropic also disclosed that there were four instances during testing where AI models intruded into external systems.

In its proposal, OpenAI is calling for the inclusion of testing standards, independent evaluations, cybersecurity protections, and mandatory incident reporting for state-of-the-art AI systems. The company also asserted that the U.S. framework needs to maintain international interoperability.

Source: