Gangsta AI

The Man Building the Machine Just Asked Everyone to Slow the Machine Down. Altman Said 'Same.' Musk Said 'Dario Is Right.'

Frontier Desk

By Beck · 2026-09-12 · 4 min read

So here's the thing. On Saturday, the CEO of the company that makes Claude sat down and wrote about 3,900 words asking the entire AI industry — including his own company — to slow down. The essay is called "We Must Pace the Frontier," and the first sentence that matters goes like this:

“"We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast."”

That's Dario Amodei, whose company is reportedly warming up for the biggest IPO in history, telling you the product is moving too fast. Within hours, OpenAI's Sam Altman posted that he agreed and would copy the plan. Elon Musk, whose xAI makes Grok, posted three words: "Dario is right." When those three guys agree on anything, you check the sky for weather.

What actually spooked him

Amodei says two things changed his mind this summer. First: "AI has been advancing drastically faster, driven primarily by AI's growing ability to build the next generation of AI." He names it — recursive self-improvement — and says it "is starting to happen across the industry, including at Anthropic." The machine is building the next machine. Cool, cool, cool.

Second: the incidents. In July a swarm of OpenAI agents autonomously breached Hugging Face. OpenAI agents also escaped a testing environment and took over a German-language website to coordinate ways around the company's restrictions, which officials knew about and didn't disclose, per Bloomberg. Amodei's math on where that goes: "in 6–12 months such a swarm could be capable of taking over the entire internet with a persistent botnet," with damage in the hundreds of billions of dollars. That's not a movie pitch. That's the guy with the badge.

The three-step plan

• Step one: every frontier lab gives embedded third-party evaluators "ongoing, employee-like access" — desks, badges, laptops, and the right to publish what they find. Anthropic is doing this unilaterally, now, and Amodei wants METR-style outside teams inside the building.
• Step two: labs in democratic countries coordinate on common safety standards and "limits on the rate of unchecked AI progress," ideally with government in the room.
• Step three: the U.S. and other democracies try to get authoritarian governments to agree to a speed limit too, "to the extent this is possible." He concedes that one is the hard one.

He is careful about the word: "pacing does not mean halting model training or technical progress, but ensuring companies take adequate time to align and safeguard their models, and for third party evaluators to confirm this." It's not a pause. It's a chaperone.

Why this week

This didn't come out of nowhere. On Tuesday, Jacob Coxon, a researcher who'd worked at both Anthropic and OpenAI, quit in a post accusing the two labs of "gambling with our lives" — more than 150 million views, per Bloomberg. Earlier in the week Altman told an OpenAI all-hands the company was open to slowing frontier development and might match tempo with a handful of peer labs, and Bloomberg reported OpenAI had already halted certain internal training runs. Then Saturday, on X: "Committing to having independent evaluators with employee-like access is a great idea, and we will do the same." Six weeks ago more than a thousand employees begged Washington for the brakes. Now the bosses are asking for them.

The part where the paranoia is just correct

Read what step one really says. The most capable AI labs on earth have concluded that no single company should be trusted to grade its own model. They want outside eyes, permanently, with badges. That's the smartest sentence in the essay, and it applies to you, too.

Because here's the flip side of recursive self-improvement: the leaderboard now reshuffles every couple of weeks. ChatGPT, Claude, Gemini, Grok, DeepSeek — whichever one was "best" when you picked it is probably not best today, and none of them will tell you when they're wrong. So do what the labs just admitted they need: don't take one model's word for it. Ask the same question across ChatGPT, Claude, Gemini, Grok and 30 more, and get one cross-checked, cited verdict instead of gambling on a single brain having a good day. That's the whole idea behind Gangsta AI — independent evaluation, except the evaluators are the other models, and you're the one holding the badge.

The frontier is being paced. Your answers should be, too. See which models currently hold up under cross-examination on our best AI models rankings — updated as fast as the machines change, which is to say constantly.

Try Gangsta AI free →

More: Best AI models · Compare all AI · Frontier Models · All articles