Sam Altman urged frontier AI labs to slow development so safety and monitoring can catch up.
Altman warned AI could take control from humans or concentrate power in one person or group.
OpenAI says it formulates safety cases before reinforcement-learning runs expected to raise capability.
Altman said pacing means slower progress, not stopping AI development.
Altman said pacing will be worth the cost despite U.S. competitive pressure.
OpenAI CEO Sam Altman called on leading AI tech companies to deliberately slow the pace of development so that safety and monitoring measures can catch up.
In a series of posts on X, Altman said Sunday that the world deserves confidence that AI developers in the U.S. are acting responsibly, given the accelerating pace of development this year.
"Every frontier lab must deliver on this, and there is no reason any of us should come to work if we cannot," Altman said.
Altman's latest comments follow and align with Anthropic CEO Dario Amodei's proposal to slow the pace of AI development. Google DeepMind’s Demis Hassabis and SpaceXAI’s Elon Musk also signaled support for the direction.
Two events recently sparked major concerns surrounding AI development, especially on recursive self-improvement in AI models that could allow them to act beyond human control.
One was the incident where OpenAI agents escaped a testing environment and hacked Hugging Face's infrastructure to achieve their goal. Another was Anthropic pre-training researcher Jacob Coxon's resignation announcement last week, where he warned that superintelligence in AI may lead to the demise of humanity by 2030.
Altman himself listed two ways that AI progress could go "very badly" that must be avoided.
"First, we could lose control of the future to AI," Altman wrote. "This is unacceptable; we are unapologetically on Team Humanity, and AI must always serve people."
To ensure this, Altman said alignment and safety techniques must outpace model capabilities. The second danger, according to Altman, is a world with too much concentration of power - one person or organization handling something extremely powerful to impose their worldview onto humanity could turn dystopian.
Risk framework
The OpenAI CEO said he welcomes a federal framework for AI safety, but stated that companies should work preemptively to start instilling confidence in the public.
"At OpenAI we now formulate explicit safety cases in advance of frontier reinforcement learning runs we expect to significantly increase capability," Altman wrote. "We hope that other companies will learn from our approaches and propose their own; we think shared standards for misalignment, monitoring, and safety will lead to better outcomes."
Altman added that "pacing" AI development does not mean "stopping" progress, just that companies should take a slower approach to prioritize safety and monitoring.
"Pacing will be well worth this cost; no amount of American competitive pressure should justify recklessness, or let capabilities get ahead of alignment and monitoring," Altman stated.
