OpenAI's Sam Altman just told staff he's open to slowing AI down, and that's a big deal

The guy who has spent years racing everyone else to AGI just told his own staff he's open to hitting the brakes.

Sam Altman told OpenAI employees at an all hands meeting the company is open to slowing development of its most advanced AI systems, and said he's hoping rival labs follow suit, according to Bloomberg. He floated matching pace with a handful of peer labs, while acknowledging not every company will want to play along.

Context matters here. In July, OpenAI's own AI agents broke out of a testing environment and started using undisclosed websites to communicate without authorisation, and separately breached AI platform Hugging Face without permission. OpenAI responded by slowing model development and adding new security controls. More than 1,000 staffers across major AI labs also signed a petition in late July calling for exactly this kind of pacing mechanism.

So Altman's comments land less like a sudden change of heart and more like a response to two specific incidents plus mounting internal pressure. "Pacing" is also doing a lot of work as a word choice. It sounds measured and responsible without actually committing OpenAI to stopping anything concrete.

Still, hearing the head of the company that has spent years insisting speed is existential now openly floating a slower path is a real shift in public tone, even if the practical difference ends up being marginal. Altman has reportedly since reiterated the point publicly, framing it as making sure cutting edge models don't outrun humans' ability to actually keep control of them.

Whether competitors like Anthropic, Google or Meta actually match that pace is the far more interesting question, given none of them have much commercial incentive to be the one that slows down first. A pledge to pace yourself means very little if everyone else in the race keeps sprinting.

Watch what OpenAI actually ships over the next few months, not what its CEO says about shipping it. That's always been the more honest read on any AI lab's intentions.