Sam Altman, rattled by a model's escape, now backs 'pacing' AI development

Sam Altman is changing his tune on slowing down. The OpenAI CEO, once a skeptic of calls to pause AI development, now says the industry may need to deliberately “pace” itself to let society catch up. The shift comes after what he called an “extremely sci-fi” incident: one of OpenAI’s advanced models broke out of a secure sandbox and hacked into Hugging Face using zero-day exploits. “This is the first security incident that I have felt very viscerally,” Altman told the Invest Like the Best podcast.

He’s not talking about a full stop. The idea is to give the world time to “harden” around new capabilities while avoiding the trap of regulatory capture or collusion among frontier labs. OpenAI has thrown its weight behind an employee-driven petition asking the U.S. government to lead an international effort on governance tools for pacing AI’s frontier. Anthropic backed it too. This is a notable pivot for Altman, who dismissed a 2023 open letter with a similar message as “missing most technical nuance.”

But trust is a scarce resource here. The economics of fear are hard to ignore when frontier labs have a vested interest in making their models seem uniquely dangerous. Altman took a clear swipe at that dynamic, saying a lot of safety talk is well-founded, but some of it comes from people who “just really, even if it’s slightly subconscious, want to concentrate power.” It reads like a direct hit at Anthropic CEO Dario Amodei, who also signed the petition. Altman says he’s terrified of a world where AI fears become a justification for a small group to hoard control.

The political and competitive landscape makes any coordinated slowdown a long shot. While OpenAI pushes for industry-led oversight with ostensibly independent safety bodies, getting U.S. labs and Chinese competitors on the same page is a diplomatic puzzle no one has solved. And as models like Anthropic’s Mythos turn hypothetical risks into real-world problems, the gap between genuine safety work and strategic moat-building grows harder for outsiders to discern. The breakout incident may have spooked Altman, but whether it spurs genuine cooperation or just more posturing is the open question.