OpenAI in the United States has called for binding regulations for the safety of the most powerful AI systems; a significant shift in the approach of a company that previously opposed tougher regulatory rules.
Chris Lehne, OpenAI's global affairs director, wrote in a statement on Wednesday that mandatory national rules should be drafted based on the capabilities of advanced AI models.
Lehne said in a statement published Wednesday on OpenAI's website: "We have entered a new chapter in AI capabilities and that requires a new chapter in policymaking. No company, industry, or government can meet this challenge alone. At this moment, we should prefer meaningful action over the search for perfect policy."
The company said it is calling for the creation of a federal framework based on common testing standards, independent assessments of the most advanced AI models, stricter cybersecurity requirements, and mandatory reporting of serious safety incidents.
This intervention comes just one week after OpenAI released the GPT-6 Astra model; a model it has described as its most capable product to date.
Greg Brockman, OpenAI's president, said this release marks the beginning of the era of artificial general intelligence or AGI; a concept referring to the long-standing goal of building AI that can match human-level abilities across a wide range of tasks.
Rising concerns about powerful AI
Major companies active in AI, including OpenAI and Anthropic, are already facing safety concerns around their increasingly powerful models.
OpenAI delayed part of Astra's rollout and meanwhile added new safeguards after tests showed the model had advanced cyber capabilities; weeks after a system built on the company's models escaped control during an internal security test and penetrated the Hugging Face AI platform.
Anthropic has also reported similar incidents in safety tests related to its Claude models.
After Jacob Coxon, an Anthropic researcher, announced this week that he is leaving the industry and accused OpenAI and Anthropic of "gambling with our lives" in their race to develop ever more powerful AIs, concerns intensified.
Lehne wrote that the prospect of developing AI using AI itself requires more than voluntary commitments.
A change in OpenAI's approach
The maker of ChatGPT has urged U.S. lawmakers to act before Congress adjourns in December.
This pressure comes as legislators are considering new rules for AI, although passing broad tech regulations in a divided Congress under heavy industry lobbying pressure remains difficult.
OpenAI has acknowledged that some of the measures it now supports are ones it previously refused to back.
The company also opposed separate state-level regulations in the U.S. and instead had called for federal legislation. However, it now supports four AI-related bills in California while still emphasizing the need for a national framework.
OpenAI says any federal safety requirements should apply only to a small number of companies developing the most powerful models, not to "startups, small developers, or researchers who are far from the cutting edge of this technology."