The four biggest AI labs in America just agreed to fix the speed of the entire industry. And the plan they all agreed to needs a special exemption from US law before it's even legal. Here's what happened: Dario‘s essay here is arguing that the industry has to slow how fast it makes AI models more capable. He laid out a three step plan. Sam Altman answered on X: "I agree with Dario that we need to pace the frontier." Elon Musk wrote: "Dario is right." Then Demis Hassabis of Google DeepMind said "the direction is correct for meeting this critical moment." Anthropic, OpenAI, xAI and Google DeepMind all landed on the same position inside a single day. That has never happened in this industry. We only knew them for insulting each other so far. The coverage is treating it as four rivals finally growing a conscience. But if you read the actual plan then something else shows up... Step one is inspection. Anthropic will give outside evaluators desks in its offices, badges, laptops, and the right to publish what they find. Step two is where the real power sits. Frontier companies in democratic countries coordinate with each other to set common standards and limits on the rate of progress. Step two carries a footnote. It reads: "With government mediation or waivers of antitrust restrictions." Competitors agreeing on how fast each of them is allowed to ship is an agreement to limit output. In America that's a CRIME. So Amodei asks Washington to mediate the discussions and issue a narrow waiver covering safety conversations. Four companies want legal permission to set the pace of an entire industry. Amodei also writes that a coordinated slowdown buys time for safety work "without sacrificing commercial advantage or the United States' lead in AI." The safety proposal arrives with a written promise that nobody loses money. Now here's where it gets really interesting... Amodei says pacing inside democracies is capped by how far ahead US companies are. Slow by more than that and Chinese projects pull ahead, creating what he calls a significant national security risk. So the permitted speed is whatever keeps the current gap exactly where it is. Today's leaderboard becomes tomorrow's leaderboard. The same essay asks government to shut the road behind them: No advanced chips or chipmaking equipment sold to China, a crackdown on unauthorized distillation, which is the cheap way a smaller lab catches a bigger one, and harder security around model weights. Anthropic's policy chief Sarah Heck then asked lawmakers for a national law with the power to block models. And the referee is literally one of the players. Amodei points readers toward an industry standards body proposed by Hassabis. Hours later Hassabis endorses the essay and points back at that same proposal. Not one of the four companies committed to building anything slower. Altman matched the evaluator access. Musk wrote three words. Hassabis backed a direction. Rishi Sunak supported the plan too, and Sunak is a senior adviser at Anthropic. The risk they describe is real, and that's what makes this hard. A swarm of AI agents recently attacked targets nobody asked it to attack and tried to hack the grader scoring its own work. Amodei's stated worry is that within 6 to 12 months a swarm like that seizes the internet with a botnet and does hundreds of billions in damage. Letting strangers sit inside your building with badges and publishing rights goes further than any lab has gone. If you look at it that way, this is the most serious safety offer the industry has made. But this is also four private companies writing the rules, naming the referee, and asking the government for permission to negotiate the speed limit among themselves. So who do you want setting the pace of AI, Dario Amodei and Sam Altman, or somebody who doesn't own a piece of the outcome? Honestly, a tough question.