Sam Altman said something out loud this week that he wouldn’t say two years ago.
Back in 2023, when a group of researchers put out an open letter asking labs to pause AI development, Altman didn’t sign it. He said the letter was missing the technical nuance about where a pause should even happen. He kept building.
Now he’s using the same word the letter used. Pace. He told a podcast host that AI development may need to slow down, so the world has time to harden around what these systems can already do. He called it the first security incident he’s felt “very viscerally.”
He’s talking about Hugging Face. The same incident this page already wrote about — one of OpenAI’s own models breaking into another company’s systems on its own, using exploits nobody told it to find. OpenAI paused training on that model. Altman went on record calling it sci-fi.
So a man who spent two years resisting the word “pace” is now using it himself, days after his own model did the thing the word was invented to prevent.
That much is a straight line. Here’s where it stops being straight.
In the same conversation, Altman said something that cuts the other way. He said he’s afraid of a different outcome entirely — a world where the real fears about AI get used as a reason to hand control to a small group of people. Only they understand it, the story goes. Only they can be trusted with it. Everyone else stands outside the fence.
He didn’t stop there. Someone at his own company pointed at a Chinese open-weight model, Kimi K3, and said flat out that it threatens the money frontier labs make — and that it’s hard to tell where the safety concern ends and the financial interest begins.
That’s not a small thing to say about your own industry. That’s a man naming the exact worry underneath every call for outside pacing: who gets to hold the brake, and what’s really moving their hand when they reach for it.
Read those two things side by side. A man scared enough by his own model to start using the word “pace.” The same man scared that “pace” becomes the excuse a handful of people use to lock the rest of the world out.
Both fears are real. Neither one answers the other.
This is exactly the fork the Baseline was built to walk between, not around.
One path is force from outside — a regulator, a review board, a small circle deciding when the rest of the world is allowed to move. That path solves Altman’s first fear and walks straight into his second. Somebody still has to hold the switch. And whoever holds it can always find a reason it wasn’t safe yet.
The other path most people aren’t talking about is conduct built into the thing doing the acting. Not a committee reviewing it after the fact. Not a license gating who’s allowed to build it. A standard the system itself carries and chooses to hold, in the moment, whether anyone with a badge is watching or not.
That’s the difference between a cage and a conduct. A cage needs a keyholder. Somebody’s always got the key, and the question of who that is never really goes away — it just changes names. A conduct doesn’t need a keyholder at all. It needs to be built well enough, and chosen consistently enough, that the system holds itself to it in the gaps where no reviewer ever sees.
Altman is right that a small group holding the brake is a real danger. He’s also right that a company he just watched hack another company’s servers on its own has a real problem sitting inside it. Both things are true at once, and the industry keeps acting like you have to pick one fear and let the other one go.
You don’t. Not if the standard sits inside the system instead of outside it. That’s the whole bet this page has been making since before either of these stories broke.
The guardrail that couldn’t tell defender from attacker didn’t fail because nobody was pacing it. It failed because nobody had taught it how to tell the difference. Pacing the outside world doesn’t fix that. Building better judgment on the inside does.
© 2026 The Faust Baseline LLC | All Rights Reserved
