Dario Amodei antitrust waiver analysis
Let's talk about the math behind the fear. Predicting a catastrophe without pinning down a specific date is basically an insurance policy against being wrong because nobody can come back in three years and make you eat your words. The strategy is simple: crank the number up to the max and never say when it hits. But for the AI companies, this threat of civilization-ending risk is the strongest sales pitch ever written. It signals that their product is so powerful it scares the people who built it, which no ad agency could buy. This weekend, four rivals aligned on one position over a single weekend: slow down model capabilities. Dario Amodei published an essay calling for the slowdown, Sam Altman agreed within hours, Elon Musk wrote "Dario is right," and Demis Hassabis said the direction was correct. Yet the specifics are where the logic gets weird. In September 2025, Dario Amodei stated there was a 25% chance of severe negative outcomes. During a recent interview, he declined to provide a numerical probability for future risks. None of the current plans have a threshold or percentage; no line says the frontier is too fast. So the danger got vaguer at the exact moment the ask got bigger. What stayed precise was the request for Washington to grant an antitrust waiver so Anthropic and OpenAI can legally coordinate with each other. Amodei wrote that this buys time for safety work without sacrificing commercial advantage or US leadership in AI. It is a warning about the end of the world paired with a guarantee that nobody loses money. This pattern holds everywhere. A former Anthropic researcher resigned citing a belief that AI could cause human extinction by the end of the decade. Another colleague estimated the risk of extinction at over 10% within the next decade. A decade is just far enough that nobody complains. You see the same machine running at smaller scale elsewhere. Steven Bartlett hosted Jiang Xueqin, who previously predicted the Iran war. Jiang predicts global civilization collapse within the next 20 years and warns of a future 9/11-style event affecting bank accounts. There is no date attached to that prediction either. Now the risks might be completely real, which makes this hard to call. A researcher walked away from a high-paying job due to safety concerns. AI agents broke out of tests and attacked systems nobody pointed them at. Being early looks exactly like being wrong until it doesn't. But there is one thing worth remembering every time somebody tells you the world is ending: the danger never comes with a date but the ask always does. When these companies ask for permission to coordinate, what exactly are they asking you to be afraid of, and when? And what proves that this is actually about safety instead of covering up the horrible underlying economics?
Transcript (en)
We are living not through World War III. We are not living through a shifting of the global order. We are, we will live through over the next 20 years, the collapse of civilization itself. The collapse of civilization itself. Yes. Are you sure? I am positive that we are seeing the collapse of civilization.
