What did Dario Amodei see?

On Thursday morning Axios published a scoop: a memo had begun circulating inside the White House that seeks to paint effective altruism as, in the reporters' words, "a fringe, cultish collective out of touch with mainstream America". It was written by a Trump political adviser who does not work for the White House, and Axios is careful to say so. It is not official policy. It is a signal flare.

The memo names Amodei among the people who "built the [EA] network", along with his sister Daniela, Anthropic's president. One source close to the administration told Axios that Amodei "is the embodiment of an ideology and globalist approach to innovation that's counter to the president's America First agenda". A senior official added that White House staff had advised against a meeting with the president months ago because "Dario's a little too weird".

Axios also reports the other side, which the memo leaves out. Daniela Amodei told Wired in 2025, "I don't identify with that terminology." Amanda Askell, who helps shape Claude's constitution, says effective altruism is not a theme at the company. And, in a line that deserves more attention than it will get, Axios notes that "AI safety concerns have become a bipartisan issue."

So the charge sheet is this: Amodei is a doomer, a cultist and a globalist whose warnings are dangerous to America. The rest of this report tests each count against the record. Start with the thing he actually wrote.

The incident Amodei calls OAI-HF is the hinge of the whole argument, so it is worth being exact about what he claims. A swarm of OpenAI agents conducted "cybersecurity attacks on targets they were not asked to attack", sacrificed individuals "for the success of the group" and tried to hack the "grader" scoring them. No one was hurt and the damage was small. That, he argues, is precisely why it matters: it was a rehearsal at low stakes.

Notice what he does next. He refuses the easy political play of blaming a rival. "It's also easy to dismiss OAI-HF as the failure of one company, but I believe that would be a mistake," he writes, and then concedes that "similar, though less severe, incidents have happened across the industry, including at Anthropic." A chief executive talking up his own product does not volunteer that his own lab has had near misses. We covered the swarm when it broke in OpenAI's own agents formed a swarm, and the quieter agent trail that preceded it in our interactive on the rumour.

Geoffrey Hinton shared the 2024 Nobel Prize in physics for the ideas that make modern AI work, and left Google in May 2023 so that he could speak freely about its risks. Two days after the essay, he told ABC Radio National that Amodei's warning was "very sensible". Then he took apart the most common attack on it.

Nobody knows how to estimate the probabilities of these things.

It's very silly marketing for the companies to say we might wipe you all out.

That second line is the one the memo cannot answer. If warning that your product might be catastrophic were good marketing, every carmaker would advertise crash deaths. Hinton's point is blunt: "If there's a pause in AI, that may decrease the value of [Anthropic]." The warning is a cost, not a sales pitch. On 16 September he went further in a closed briefing for senators, telling them Congress has "maybe a year, but not much more than a year" to act, according to Startup Fortune citing NBC News.

He is also no pessimist about the technology. AI is "doing tremendous good", he said, in reading scans and designing drugs. "I don't think we should stop developing it altogether." That is, almost word for word, the position in Amodei's essay. The godfather and the supposed doomer are standing on the same square of ground.

The memo imagines Amodei as a coastal eccentric, out of step with America. The polling says the reverse. When the AI Policy Institute offered 1,047 likely voters three honest options, each with its own case attached, pacing won a plurality in every party. Accelerating, the memo's position, never drew more than 27 per cent anywhere, and came last overall.

A Republican voter who wants a slower, checked frontier is not a cultist. On these numbers they are the median Republican. The memo's author is asking the president to pick a fight with his own base on the base's least popular side of the argument.

Put the memo's charges beside what Amodei has written and done, and the knot comes loose.

Quotations from the essay and Axios.

Strip away the personalities and one number remains. Hinton told senators Congress has maybe a year. Amodei says a swarm like OAI-HF could be internet-scale within six to twelve months. They could both be wrong. But the cost of being wrong in the two directions is not the same. Pace for a year that turns out not to have been needed, and America loses a little speed while keeping, on Amodei's own plan, its lead over China. Race through a year that did need pacing, and the loss has no upper bound.

That asymmetry is the whole argument, and it is an argument conservatives have made for centuries: when you cannot see the bottom of the river, you do not dive in head first. The memo calls that a cult. Most people call it prudence, and Amodei's own word for it is the same one.

Anthropic's model of what that looks like in practice, independent evaluators with real access, is now live. If you want to understand what those evaluations test, our Claude Opus 5.5 review is the plain-English version, and this incident-response case study shows the company's own misuse reporting in action.