Sam Altman Reportedly Told Staff OpenAI Could Slow AI-Agent Development With Other Labs

Sam Altman told OpenAI employees the company could slow development of AI agents, potentially alongside other AI labs, according to a report. The proposal puts a practical condition on any meaningful pause: OpenAI can alter its own pace, but a coordinated slowdown requires competitors to agree.

Altman reportedly raised the possibility at a company-wide meeting and acknowledged that some labs might not be willing to participate. That caveat is central. The debate is not simply whether one company should pause work, but whether safety-driven restraint can hold when companies developing advanced systems are competing with one another.

A stated policy becomes a current choice

OpenAI had already publicly set out a standard for intervening in development. In its September 6 research publication, the company said it would slow or stop development or deployment when moving ahead would create an unacceptable safety risk. The post includes a section explicitly devoted to pacing model development.

That policy was not purely hypothetical. OpenAI said it paused reinforcement-learning training on its latest models intended for deployment after a recent incident, while it hardened research environments, expanded monitoring, and red-teamed those environments. Some work resumed under stronger controls while other workloads remained paused.

Companies should be “coordinating to slow down future development as needed.”

Jakub Pachocki, OpenAI chief scientist

Pressure from inside the field

Pachocki’s call for coordination came amid public concern from people working in and around leading AI labs. More than 1,100 staff members at major technology companies, including OpenAI, Anthropic, Google and Meta, signed a petition urging deliberate pacing of advanced AI-agent development.

Jacob Coxon, a researcher who worked at Anthropic and OpenAI, alleged that both companies were racing toward self-improving superintelligence without acting responsibly. That is Coxon’s assessment, not evidence that such a system exists today. Anthropic staffer Drake Thomas and former Google DeepMind researcher Anna Wang also publicly called for stronger safety barriers or a slower pace of development.

The limits of a voluntary slowdown

Altman’s reported remarks leave key terms unresolved: what safety threshold would trigger a broader slowdown, which work would be affected, and whether other labs would accept the same limit. OpenAI’s existing policy describes action when it judges risk unacceptable; it does not establish an industry-wide agreement.