Anthropic CEO Dario Amodei is urging both U.S. policymakers and industry to wake up to the idea that powerful AI constituting a “country of geniuses in a data center” may pose “the single most serious national security threat” faced by humanity in a century. (Amodei did not elaborate on which particular national security threat from a century ago he was referring to.)
In a new essay, “The Adolescence of Technology,” Amodei argues that “powerful AI” systems may materialize as soon as the next one to two years. The essay, published Monday on Amodei’s website, darioamodei.com, follows on from “Machines of Loving Grace,” which explored the revolutionary benefits of powerful AI.
In the essay, Amodei predicts that by about 2027, cluster sizes, or the interconnected computing resources grouped together to train or power AI, will allow for the running of millions of AI instances, each operating at superhuman speed. Imagine powerful AI as a “country” with the knowledge of 50 million Nobel Prize winners, he writes, possessing the kind of brainpower that would put the world at risk merely by existing.
“It is clear that, if for some reason it chose to do so, this country would have a fairly good shot at taking over the world,” Amodei writes. For instance, it could grow hostile and take over the world itself, or could help an existing bad actor to do so. Or its advanced capabilities could disrupt the global economy and cause mass unemployment. Finally, this “country” may bring about destabilizing effects indirectly owing to the new technology and productivity advances it will bring about.
Even if a hostile AI country doesn’t get humans to do its bidding, it could still affect the world by building an army of robots or taking over tech-connected infrastructure and devices, Amodei suggests.
Anthropic did not immediately respond to Fortune’s request for comment.
Amodei’s “country of geniuses” metaphor builds on his comments at the World Economic Forum meeting in Davos, Switzerland, earlier this month, where he repeated his controversial prediction that AI is advancing so rapidly that it will replace the work of software engineers within a year and eliminate half of all white-collar jobs within five years.
“I think we’re going to be surprised at how the exponential turns upward. The whole thing about exponentials is, you know, it looks like it’’s going very, very slowly. It speeds up a little bit, and then it just zooms past you. And I think we’re on the precipice,” Amodei told Bloomberg at Davos.
Amodei isn’t alone in seeing the exponential impact beginning, starting now. Goldman Sachs economist David Mericle, who has previously coauthored research about how AI is part of the equation ushering in an era of “jobless growth” in the U.S. economy, predicted on Monday that net job losses in the most AI-exposed industries will “increase meaningfully” in 2026.
Despite his bleak conclusions, Amodei rejects “doomerism” and the idea that an AI catastrophe is inevitable. To counter the potentially disastrous outcomes of powerful AI, he said Anthropic has employed a post-training method called Constitutional AI for its large language model, Claude.
This post-training method tries to steer a model’s behavior using a central set of values and principles instead of a long checklist of forbidden requests. Anthropic argued that this values-first approach is designed to teach Claude to be “a good AI.” By the end of 2026, Amodei said Anthropic wants to train Claude so it almost never goes against the spirit of this “constitution.”
“This is like a child forming their identity by imitating the virtues of fictional role models they read about in books,” he said.
Still, while he sees Constitutional AI as a step forward, Amodei also made clear that he doesn’t see it as a catchall solution. He encouraged more transparency laws like those passed in California and New York that require AI developers to disclose how their systems are built and trained.
Ultimately, Amodei writes that he views the current era as a test of human character, and urges the world to meet head-on the potential challenge of a fast-approaching, powerful AI.
“Humanity is about to be handed almost unimaginable power, and it is deeply unclear whether our social, political, and technological systems possess the maturity to wield it,” he writes.
