OpenAI published a post on Wednesday urging governments to impose mandatory safety requirements on frontier AI systems, stating that there's a limited window of time to establish such rules.
"We've reached a new chapter in AI capabilities, and that demands a new chapter for AI policy," OpenAI Chief Global Affairs Officer Chris Lehane said. "Policymakers face an analogous moment: a closing window to establish durable safeguards before AI capabilities outpace the institutions responsible for governing them."
OpenAI's latest post comes a day after Jacob Coxon, a pretraining researcher who has worked at both Anthropic and OpenAI, wrote a post on X that quickly went viral, drawing 138 million views and triggering intense debate over the risks of accelerated AI development.
Coxon, while announcing his resignation from Anthropic, warned that the two AI labs are racing toward self-improving superintelligence without considering the risks to civilization. Coxon stated that it is a shared belief among AI developers that the technology could kill humanity by 2030.
Lehane noted that fully autonomous recursive self-improvement in AI is not happening today, and that companies should not pursue it unless there are proper safeguards.
However, Lehane said AI has already accelerated human capabilities in many fields, and this should be managed through meaningful human control and globally shared safety bars that can curb or stop growth.
"The more powerful the technology becomes, the stronger the surrounding safeguards must become," the post said.
Urging Congress
In its latest post, OpenAI said it seeks to collaborate with Congress on developing rules in the U.S. covering testing, independent assessment, cybersecurity, incident reporting, and measures to track progress toward AI's recursive self-improvement.
Until a nationwide rule comes into effect, OpenAI said it will continue to support state-level measures, announcing its support for four bills in California - SB 813, AB 1405, SB 1119, and AB 1864 - which cover assessing risks, independent audits, accountability requirements, and physical safeguards against AI-led threats.
"Some of these bills we did not endorse in the past, and are now supporting after reconsidering in light of the recent jump in capabilities we have seen," Lehane said. The company has previously expressed support for California's SB 53, New York's RAISE Act, Illinois' SB 315, and independent audits in the frontier safety legislation under consideration in Massachusetts.
The post also called for industry standards for monitoring AI development, especially for hints of misalignment that involve a model pursuing a goal against human intent or boundaries without having malicious intent.
The OpenAI executive said monitoring must be tied to clear disclosure requirements for companies, requiring them to promptly notify affected parties when their models bypass another organization's security controls without authorization. Such an incident occurred in July when OpenAI agents escaped a testing environment and hacked Hugging Face's infrastructure to achieve their goal.
OpenAI is currently developing a framework for reporting misalignment incidents, and said the company hopes to utilize it to support broader federal policy development.
"Models, research, and technical expertise move globally, and failures involving the most capable systems could have consequences far beyond the country in which they are developed," OpenAI wrote. "The United States needs to establish credible standards at home if it is going to lead internationally - and we are increasingly convinced that compatible international standards will be necessary."