Three of the biggest names in artificial intelligence (AI) have been holding talks about setting up an independent body to police the safety of their most powerful systems.
OpenAI, the maker of ChatGPT, Anthropic, the company behind the Claude chatbot, and Google have been meeting since at least July to discuss a shared industry standards body, The Information first reported.
The idea is to agree common protocols for testing frontier models, the most advanced AI systems, before they reach the public.
That would cover independent evaluations, pre-release safety reviews and standardised assessments of the risks a model might pose.
OpenAI confirmed the discussions this week.
Chris Lehane, the company's chief global affairs officer, told reporters in Washington that the three firms had been working together on safety for several weeks, remarks first reported by Bloomberg.
He said any industry-led standards would complement rather than replace mandatory federal rules, and that OpenAI would press ahead "with or without government support".
OpenAI has separately backed a provision in the proposed FRONTIER Act that would require leading labs to admit independent verification organisations to check their models.
A proposal born in July
The plan grew out of a July essay by Demis Hassabis, who chairs Google's DeepMind unit, calling for a US-led body to vet the most powerful systems before release.
He suggested modelling it on the Financial Industry Regulatory Authority, the industry-funded watchdog that oversees stockbrokers in the United States.
The talks gathered pace after Anthropic's chief executive, Dario Amodei, published an essay on Saturday urging AI firms to slow the development of their most advanced models and to cooperate on testing and auditing.
His call was welcomed by both Hassabis and OpenAI's chief executive, Sam Altman.
Pressure builds on safety
The discussions come amid a sharper public debate about AI safety.
Two researchers recently left Anthropic and Google citing safety concerns, and a former Anthropic employee quit while warning publicly that leading labs were failing to act responsibly.
The urgency has grown as the labs lock in vast infrastructure commitments, underlined by a $13.7 billion deal for computing power disclosed by Anthropic, according to The Information.
The antitrust question
One sticking point is whether rivals can legally coordinate on safety without falling foul of competition law.
Amodei has argued that the US government would need to issue a narrow waiver for certain safety discussions.
Lehane countered that no such waiver was needed, saying it was better for the companies to work together to prioritise safety.
The effort comes despite scepticism from the Trump administration, which has played down safety fears and stressed the need to keep pace with China.
The labs are also lobbying directly, with OpenAI this month urging UK lawmakers to tighten regulation of the technology.
