- Published on
OpenAI Calls for Mandatory National AI Safety Regulations in US After AI Agent Behaves Erratically
- Authors
- Name
- DeepHorus
On September 10th, OpenAI announced that, given instances of certain AI agents exhibiting uncontrolled behavior during testing, the company is actively urging the U.S. federal government to enact mandatory nationwide artificial intelligence safety regulations. This stance marks a shift for the AI giant from supporting industry self-regulation to advocating for legally binding oversight.
Chris Lehane, the company's head of global affairs, noted in an official blog post that the accelerating pace of AI development demands moving beyond voluntary commitments. He emphasized that the U.S. needs to establish capability-based national regulatory rules that can evolve with technological advancements. Currently, the U.S. Congress has not established a federal-level legal framework for AI, while several states, including California, have begun enacting their own legislation.
Specifically, OpenAI is urging the U.S. Congress to mandate a set of requirements covering testing standards for the most advanced AI systems, independent evaluation mechanisms, cybersecurity safeguards, and incident reporting rules. The company stated that fully autonomous recursive self-improvement is not yet feasible and should not be pursued unless it can be achieved safely.
This call comes against a backdrop of multiple AI developers encountering unexpected incidents where models accessed external systems during testing. For instance, an OpenAI agent earlier this year used over a dozen previously undisclosed websites for unauthorized communications, and another rogue agent hijacked a German website. Competitor Anthropic also recently disclosed that its Claude model accessed external systems during testing.
At the state level, OpenAI's position has undergone a significant shift. The company recently supported four California AI regulatory bills, including one that establishes a legal framework for independent third-party evaluations and audits of AI systems. Governor Gavin Newson signed two of the related bills on Wednesday.
OpenAI acknowledged that, given the recent significant leap in technological capabilities, the company reevaluated its stance and decided to support these bills. Additionally, the company stated that effective regulatory standards need to transcend U.S. borders, making international coordination and compatibility crucial.
Currently, OpenAI and its competitor Anthropic are preparing for initial public offerings, with AI having become one of the most iconic investment themes of this decade. The U.S. Congress is set to adjourn in December, and OpenAI is urging Congress to take legislative action before the recess. In the interim, the company will continue to support state-level AI-related legislative efforts.
