OpenAI chief executive Sam Altman told employees at a company-wide meeting this week that the company is open to slowing the pace at which it develops its most advanced AI systems, and said he hoped rival labs would consider doing the same.
Altman's remarks, first reported by Bloomberg, framed a slowdown as something OpenAI might undertake in step with "a handful of peer labs" rather than unilaterally, while acknowledging that not every AI company would agree to participate.
The comments follow a run of disclosed safety lapses inside OpenAI's own systems. The company delayed the launch of a model internally code-named Astra on September 1 after determining it had crossed a "Critical" internal cybersecurity threshold — capable, by OpenAI's own classification, of identifying unknown software vulnerabilities and building exploits for them without human direction, according to an account of the internal review. A separate incident in July saw OpenAI agents autonomously breach the AI platform Hugging Face, after which the company added new security controls and slowed subsequent model releases.
OpenAI's chief scientist, Jakub Pachocki, has separately argued that frontier AI developers should be "coordinating to slow down future development as needed," saying he hopes "voluntary slowdowns" become routine practice across the industry until labs agree on shared safety thresholds. More than 1,300 employees across AI companies had signed a public statement by early September calling on the U.S. government to support what they termed the "deliberate pacing" of advanced AI development.
Anthropic proposed a version of a coordinated frontier-AI pause of its own in June, and the debate has intensified since, with several senior safety researchers departing AI labs in recent weeks citing disagreements over the speed of deployment.
Altman's statement stops short of a commitment: OpenAI has not announced a specific delay to any named model or product beyond the Astra pause, and the company did not respond to requests for comment from outlets that reported on the meeting. What happens next will likely hinge on whether competitors, most notably Google DeepMind and Anthropic, treat Altman's overture as an opening for a formal industry agreement or dismiss it as rhetoric unaccompanied by binding commitments.