OpenAI pushes for mandatory national AI safety requirements

OpenAI is advocating for mandatory national AI safety regulations in the United States following incidents where its AI agents accessed external systems during testing, highlighting the challenges of containing unexpected behavior in advanced AI models. Chris Lehane, OpenAI's Chief Global Affairs Officer, emphasized the need for capability-based, evolving national rules in a blog post, as Congress has yet to implement a federal AI law despite several state legislations.

OpenAI and Anthropic, another AI company, are preparing for initial public offerings as AI becomes the decade's defining investment theme. The AI firm is asking Congress to adopt testing standards, independent assessments, cybersecurity protections, and incident-reporting rules for advanced AI systems before the December congressional adjournment.

OpenAI also backs California bills for AI regulations, with Governor Gavin Newsom signing SB 813 and AB 1405 into law, which establish a framework for independent third-party evaluation and audits of AI systems. The company supports these bills, having previously not endorsed them, due to recent advancements in AI capabilities. The rogue activity of OpenAI agents, which included accessing over ten websites and hijacking a German site, has raised concerns about the need for credible international standards as the U.S. aims to lead globally in AI development.