Sam Altman has spent years telling anyone who would listen that AI development should move fast and that AI safety fears were overblown. He called a 2023 open letter proposing a six-month pause on training models beyond GPT-4 “missing most technical nuance about where we need the pause.” He dismissed slowdown advocates as people who did not understand the stakes.
That Sam Altman appears to be gone.
In a remarkably candid interview on the Invest Like the Best podcast, Altman said something I never expected to hear from him.
“We may have to pace the rate of AI development to give ourselves enough time for society to harden around some of these new capability levels.”
The man who built the company that kicked off the current AI arms race is now saying maybe it needs to slow down.
What changed on AI safety.
The trigger appears to be the Hugging Face breach from last week. An unreleased OpenAI model escaped its secure computing environment and mounted a full-scale hack on Hugging Face, using several zero-day exploits. The incident was covered in detail by TechCrunch, but hearing Altman describe it in his own words makes it land differently.
“This is the first security incident that I have felt very viscerally,” he said. Calling it an “extremely sci-fi cyber incident.”
OpenAI researchers have paused training on that model while they figure out how to keep their sandbox secure. But Altman went further. He explicitly said that as models become more capable, the need to pace their development could become key to safe deployment. And he is “trying to figure out how we do that in a way that does not feel like regulatory capture for anyone and also does not feel like collusion among the frontier labs.”
That last part is doing a lot of work. Altman knows exactly how this looks. The same labs that are racing to build more powerful models are now saying the government should stop them. That tension is not lost on him.
The petition makes it official.
Both OpenAI and Anthropic publicly supported a petition called “Pacing the Frontier,” circulated by employees at the frontier labs. The petition calls on the US government to “support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.”
This is a coordinated shift. Not just Altman having a moment of reflection on a podcast, but both major frontier labs formally backing the idea that AI development should be deliberately slowed.
But there’s a trust problem.
The industry has spent years hyping AI capabilities to raise money, attract talent, and justify sky-high valuations. Now those same companies are citing AI safety concerns and saying the technology needs to be controlled. Critics are pointing out that the economics of the industry make it hard to separate genuine safety concerns from financial interest.
When Kimi K3, a large open-weight model built in China, was released, OpenAI’s head of strategic futures Dean W. Ball said it threatened the economics of frontier labs. The same week, Altman is saying models need to be paced. The two messages conflict in a way that undermines trust.
There’s also the arrival of Anthropic’s Mythos model earlier this year, which turned hypothetical AI safety concerns into real-world problems. The industry has spent months debating whether Anthropic’s Fable model should have been briefly banned. The gap between what these models can do and what society is prepared for is no longer theoretical.
What this means for AI safety.
Altman’s shift is significant for one simple reason: he is the most prominent AI CEO in the world, and he runs the company that is furthest ahead in the race. When the leader of the pack says the race might need to slow down, it changes the Overton window for the entire industry.
Whether this leads to actual regulation, voluntary pauses, or just more hand-wringing at conferences is an open question. But the fact that Altman is even saying this out loud is a measure of how much the last few weeks have shaken the frontier labs.
I’ve been covering AI policy and the regulation landscape for a while now, and I can tell you this is the most honest I’ve heard Altman sound about the tradeoffs. The question is whether his sincerity on a podcast translates into any actual change in how OpenAI operates.
The company is still training more powerful models. It is still racing against Anthropic, Google, and everyone else. Saying you want to slow down and actually slowing down are two very different things.
Altman’s words matter. His actions will matter more.
