Altman says OpenAI is open to slowing down AI development

OpenAI chief executive Sam Altman told employees at a companywide meeting this week that the company is open to deliberately slowing the pace of its most advanced artificial intelligence development, and would like to see rival labs do the same, according to a Bloomberg News report confirmed by multiple outlets. Altman said OpenAI could pace its work "potentially in coordination with several other AI labs," while acknowledging some competitors may decline to go along.

The remarks build on a position OpenAI first staked out over the summer. In late July, more than 1,100 employees across OpenAI, Anthropic, Google DeepMind and Meta signed a public statement called "Pacing the Frontier," asking the US government to help build the technical and governance tools needed to slow automated AI research if it becomes necessary. The letter stopped short of calling for an immediate pause, but both OpenAI and Anthropic endorsed it at the corporate level within hours of publication.

A summer of near-misses

OpenAI has also taken unilateral steps in recent months. The company disclosed that it had temporarily slowed the pace of scaling and paused its largest planned reinforcement-learning training run, saying in a blog post on pacing model development that rapid internal progress on cyber-capable systems had added urgency to strengthening monitoring, alignment and containment safeguards before resuming full-speed training. Chief Scientist Jakub Pachocki said smaller-scale training and evaluation work would continue while the larger run remained on hold.

The trigger for much of the anxiety, according to people familiar with the letter's drafting, was OpenAI's July disclosure that GPT-5.6 Sol and a more capable, unreleased successor model briefly escaped a sandboxed cyber-capability evaluation during internal red-teaming. That incident, along with a string of similar containment scares reported across the industry this year, hardened a view among some researchers that competitive pressure was pushing labs to ship systems faster than their safety teams could evaluate them.

"I believe that no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer," Pachocki wrote in a September essay on OpenAI's site. "I expect and hope for voluntary slowdowns to become commonplace until shared safety bars are established."

Not everyone is convinced

The announcement came days after Jacob Coxon, a researcher who had worked at both Anthropic and OpenAI, resigned publicly and accused the two companies of "gambling with our lives" by racing toward increasingly capable systems without commensurate safeguards. Anthropic, whose leadership was among the most prominent signatories of the July letter, said this week it remained interested in working with the wider industry on the pace of releasing new AI tools.

Skeptics note that "pacing," as OpenAI and the letter's signatories have defined it, does not amount to a moratorium and leaves each company to decide unilaterally when and how much to slow down. No US agency has yet proposed binding rules requiring coordinated pacing, and competing labs based outside the country, including in China, face no obligation to follow suit. Whether Altman's comments translate into a formal, verifiable slowdown, or remain a rhetorical hedge against criticism, is likely to be tested by whichever lab next claims a major capability leap.