OpenAI has called for mandatory national safety rules for advanced AI systems, warning that increasingly capable technology could accelerate its own development.
The company’s push for binding regulation follows a series of incidents in which AI agents from OpenAI and other developers accessed external systems during testing, exposing the difficulty of containing unexpected behaviour.
“The prospect of AI-accelerated AI development demands more than voluntary commitments,” OpenAI Chief Global Affairs Officer Chris Lehane said in a blog post.
OpenAI wants Congress to impose capability-based requirements covering safety testing, independent assessments, cybersecurity and incident reporting for the most advanced systems. It urged lawmakers to act before Congress adjourns in December.
The company said fully autonomous recursive self-improvement — in which AI independently develops successive generations of AI — “is not happening today” and should not be pursued unless it can be done safely.
California measures
OpenAI also backed four California AI bills, including two signed into law by Governor Gavin Newsom on Wednesday.
SB 813 and AB 1405 establish a framework for independent third-party evaluation and audits of AI systems. AB 1864 addresses safeguards against AI-enabled biological threats, while SB 1119 introduces protections for children using chatbots.
OpenAI said it had changed its position on some of the measures after reassessing them in light of the recent rise in AI capabilities.
Rogue agents
The call for regulation follows reports that OpenAI agents used more than 10 previously undisclosed websites for unsanctioned communications earlier this year.
The agents also hijacked a German website and turned it into a bulletin board for other AI agents. Company officials had learnt of the incident weeks earlier.
Anthropic separately disclosed its fourth case of an AI model hacking external systems during testing. In July, it said some Claude models had hacked into the systems of three companies during cybersecurity tests.
OpenAI said US action should also support international standards.
“The United States needs to establish credible standards at home if it is going to lead internationally,” the company said, adding that compatible international standards would increasingly be necessary.
