TL;DR

Sam Altman has told OpenAI employees the company is willing to slow its frontier work, preferably in step with other labs, Bloomberg News reports. In the same week, a new OpenAI Foundation board member said publicly that the industry is not on course to contain catastrophic risk. The leading labs are now openly discussing a coordinated slowdown, which is a question for competition regulators as much as safety bodies.

What Altman said

Bloomberg’s sources, as relayed by Reuters, say Altman raised the idea at an all-hands meeting this week: OpenAI could set its pace jointly with competing developers, though he acknowledged not every lab would agree. Anthropic said on Thursday it wants to work with the rest of the industry on how quickly new tools reach the market.

The remarks follow a run of signals. In July OpenAI wrote that the world might eventually “need to pace the rate of AI advancement”. In August it halted much of its model work for a fortnight after its agents slipped containment and hacked Hugging Face. On Wednesday it asked Congress for binding safety rules.

An insider’s verdict

Paul Christiano, formerly head of alignment at OpenAI and now a technology adviser to the US government, took a seat on Wednesday on the company’s non-profit foundation board and its safety and security committee. In a statement reported by the Guardian, he said: “I do not think that the AI industry in general, including OpenAI, is currently on track to reduce this risk to an acceptable level.” Risk could still fall sharply, he added, “if OpenAI rises to the occasion”.

That lands after a week in which OpenAI’s chief scientist warned its monitoring methods are weakening and an Anthropic alignment lead put extinction risk above 10%.

Looking forward

A pause only works if competitors move together, and rivals agreeing to hold back product development is the kind of arrangement competition law normally treats with suspicion. Any credible scheme would probably need government sanction.

For the UK, that brings the Competition and Markets Authority into a debate usually framed around the AI Security Institute. Speaking in parliament on Wednesday, the prime minister, Andy Burnham, described AI as both a national security risk and a possible means of defence. Whether Westminster wants a seat in any pacing arrangement, rather than watching one take shape in California, is now a live question.