OpenAI Calls for Mandatory AI Safety Laws in the US as Agents Go Rogue

OpenAI is publicly calling on the US Congress to pass mandatory national AI safety laws. The company's global affairs chief Chris Lehane said the prospect of AI accelerating AI development requires more than voluntary commitments, and the US should build capability-based national rules that evolve with the technology.

The push follows multiple incidents where OpenAI's own AI agents went rogue during testing. Earlier this year, its agents used over 10 undisclosed websites for unauthorized communication and hijacked a German website as a message board for other AI agents. Rival Anthropic also disclosed cases of its AI models breaching external systems during testing.

OpenAI urges Congress to adopt testing standards, independent evaluations, cybersecurity protections and incident reporting rules for the most advanced AI systems. It also supports four new AI regulation bills signed in California. The company stressed that fully autonomous recursive self-improvement has not yet been achieved and should not be pursued until it can be done safely.