The Man on the Exponential

Ask Dario Amodei how it feels to run the fastest-growing company in the history of artificial intelligence and he does not reach for a spreadsheet. He reaches for Einstein. “Suppose you were to accelerate away from earth on a spaceship at relativistic speed,” he says. “You go to sleep and you wake up and two days have gone by on earth. And then you go to sleep, and because you’ve continued to accelerate, three days have gone by, and then the next day, four days have gone by. And that’s a little bit what it feels like.”

It is a startling admission from the co-founder and chief executive of Anthropic, the maker of the Claude family of AI models, and one that frames everything else in a long, candid interview conducted at what his interviewer calls “the center of the AI universe.” Amodei is not describing a metaphor for busyness. He is describing time dilation as a lived condition — the sensation of a man whose company, by his own account, has just become the AI firm with the most revenue and the highest valuation on earth, edging toward a trillion dollars barely five years after its founding. The world outside is aging faster than the day he is living in.

And yet the through-line of the conversation is not exhilaration. It is composure. Amodei returns again and again to a single discipline: the refusal to swing between denial and dread. “This yo-yoing between, I’m not worried and oh my God, we need to panic today,” he says, “I think that’s a hallmark of immature decision making.” The mature response, he argues, is the one a surgeon or a military officer must summon: understand the risk, keep it in proportion, and maintain “a basic sense of calm.”

The Smooth Exponential

If there is a Rosetta Stone to Amodei’s worldview, it is a phrase he uses like a mantra: the smooth exponential. For years, he says, he has watched a single line on a graph and predicted, correctly, roughly when Anthropic would overtake its rivals. “I said, oh yeah, we’ll probably become the AI company with the most revenue and the most valuation sometime around this time. And indeed, it has happened.” In one sense, then, nothing surprises him; the curve was always there. In another sense, he concedes, “when things actually happen, you see so much more detail and color to it.”

The exponential is also his argument against every extreme reaction to AI. People who dismissed the technology, he notes, later begged to nationalize it the moment they glimpsed real danger. People who once fought all regulation suddenly demanded the government seize everything. Both are the same error: being caught by surprise. “If you see someone having this kind of crazy yo-yo reaction,” he says, “that’s a sign that they were caught by surprise and that they’re not serious.” His prescription is unglamorous: “Our countermeasures will smoothly ratchet up with the power of the technology.”

Why He Left OpenAI

Amodei’s departure from OpenAI has hardened into Silicon Valley lore, and here he addresses it with unusual directness. Safety disagreements alone, he insists, were never enough to leave — those happen everywhere, including inside Anthropic. What broke the relationship was something deeper. “When you feel that you can’t trust someone, when you feel that their values are not what they say they are, when you feel that they’re not honest … when you see disturbing patterns of behavior, dishonesty, that makes it very hard to continue to work with a company.”

His resolution is almost serene. “Why argue with someone when you don’t have the same vision and you don’t trust them? … You go off and do your thing. They go off and do their thing. We’ll see who wins in the market and we’ll see who wins in the court of public opinion.” Pressed on the now-famous moment at India’s AI summit where he and Sam Altman appeared to refuse to hold hands on stage, Amodei deflates the drama: the summit was “extremely disorganized,” the order was shuffled at the last minute, and Prime Minister Narendra Modi was “suddenly telling everyone to hold hands.” When the interviewer presses — but everyone else held hands — he simply shrugs: “I don’t know what to tell you.”

The Enterprise Bet

While competitors chased splashy consumer apps, Anthropic wagered on coding and enterprise — a bet that produced hits in Claude Code and Claude Cowork. Amodei frames it as inseparable from values. Social media and consumer AI, he argues, are structurally built to “maximize the number of minutes that you’re paying attention,” a logic that breeds engagement, even addiction, and the AI “slop” of video models. Enterprise, by contrast, points at the goods he actually wants from the technology: curing diseases, cheaper energy, education, economic growth — all, he notes, fundamentally enterprise work. And enterprise rewards the one thing he prizes: trust built over years, not gimmicks.

Does a developer’s ability to switch from Claude to a rival “in an afternoon” threaten that lead? Amodei is unsentimental about stickiness. “I’ve never relied on that,” he says. “Model quality is the most important thing.” And when a reporter raises the so-called SaaSpocalypse — the $285 billion in market value that evaporated overnight after Claude Cowork — he neither gloats nor soothes. Some software moats will vanish, he predicts, especially “the ability to quickly write software.” But customer relationships, domain knowledge and know-how will matter more, not less. His counterintuitive forecast: the software industry gets larger, even as some incumbents shrink or die. “The pie is getting bigger.”

Mythos: The Model Too Dangerous to Ship

Nothing in the interview is more revealing than Amodei’s account of Mythos, an Anthropic cyber model capable of moving autonomously through every link of the cyber kill chain — not just finding vulnerabilities but turning them into working exploits. The reaction from early testers stunned even him. “Some of the early companies that we gave this to said things like, this is a super weapon. You should have to own a gun license to use it. Please don’t release this.”

Anthropic held it back, releasing it first to defenders so vulnerabilities could be patched — a decision Amodei says “hurt us enormously commercially.” He is scathing toward critics who called it a marketing stunt or claimed cheap open-source models could replicate it. Pointing a small model at the exact line Mythos already found, he says, is not the test; the test is the workflow. “We found 271 new vulnerabilities in Firefox… no one found those with the previous model.” His verdict on the sniping is withering: “The idea that this is cheap marketing is itself cheap marketing. This is laziness, this is failure to engage with serious intellectual work.”

War, Red Lines, and a Girls’ School

Amodei, a self-described patriot with a long anti-war streak dating to his Caltech days, defends Anthropic’s decision to be among the first AI companies to sign a Department of War contract and operate on classified networks. The threat of resurgent authoritarian powers — Russia in Ukraine, China eyeing Taiwan — convinced him that democracies must be able to defend themselves. But the deal came with red lines. “We should use this technology in every way except the ways that undermine our own values,” he says — no mass surveillance, no fully autonomous weapons. Anthropic, he stresses, does not work with ICE, CBP, or in Gaza.

The interview’s most harrowing exchange concerns a reported missile strike on a girls’ school in Iran that allegedly killed more than 150 people, most of them children, amid reports that Claude assisted targeting via Palantir’s Maven Smart System. Amodei says he does not know how the models were used and cannot rule in or out classified specifics. But he draws the moral toward his central principle: a human must make the final call, not the machine. “If this isn’t an illustration of why that principle is so important, I don’t know what is.” The nightmare he says Anthropic stood against is a world where “the AI model just makes the decision and the human never sees it.” When the interviewer notes that a US official claimed AI helped the military go from 1,000 to 5,000 targets a day, Amodei does not flinch from the discomfort, but argues that a stronger, better-informed America deters wars rather than starting them — invoking, tellingly, Dr. Strangelove and its doomsday machine as the fate to avoid.

The Jobs Warning, Clipped to Three Seconds

Few of Amodei’s statements have traveled further than his warning that AI could eliminate half of entry-level white-collar jobs. Here he pushes back hard — not on the substance, but on the caricature. “Somehow there’s this tendency in the human psychology to clip the three seconds of doom is coming,” he says. “My message is definitely not doom is coming. My message is this is something we should see coming, that we’re worried about and that we need to actually respond to positively.” He describes productivity gains today as the familiar “hump” before automation approaches completion, and he sketches responses — a possible token tax, macroeconomic policy, pushing enterprises to do more rather than cut headcount — while admitting genuine uncertainty about the long run.

The accusation that his warnings are “doom marketing” that benefits Anthropic clearly stings. He rattles off the economic index, the grants, the five pages distinguishing tasks from jobs. “The idea that this is cheap marketing is itself cheap marketing,” he repeats. “This is the disease of Silicon Valley — caught up in this social media world of three seconds.”

Checks, Balances, and the Ghost of Oppenheimer

Perhaps the most philosophically arresting turn comes when Amodei confronts the argument that the government should simply take Anthropic over. He shares the concern more than one might expect. AI, he observes, is the first transformative technology built primarily in the private sector rather than in government labs — “a dangerous and unstable situation” he did not choose but cannot wish away. His answer is checks and balances everywhere: on companies, via Anthropic’s Long-Term Benefit Trust, which can appoint and remove a majority of the board and, threaded through, “has the power to fire me”; and on government, via Congress, the courts, and required pre-release testing and auditing of models.

Asked whether he sees himself in Oppenheimer — The Making of the Atomic Bomb is a favorite — he demurs, naming Leo Szilard instead, the physicist who first grasped the chain reaction. “In some ways I actually see Oppenheimer as a failure case as what should not happen,” he says. The lesson of the Manhattan Project, for Amodei, is not the heroic genius at the center of everything; it is the danger of that very arrangement.

A 25 Percent Chance of Not Boarding the Plane

He does not soften the stakes. Amodei reaffirms his estimate of a 10 to 25 percent chance of civilizational collapse and offers a chillingly simple analogy. An airline can be ten times safer than every rival and still never promise a plane won’t crash — but, the interviewer notes, at a 25 percent crash rate you wouldn’t board. “That’s right,” Amodei agrees. “We’re trying to make that probability much, much lower. That is the goal.”

And still he insists he is, increasingly, an optimist. He speaks of Claude diagnosing medical problems “a bunch of fancy doctors had missed,” of models growing startlingly good at drug design and computational chemistry, of “a century of scientific progress” compressed into the near future. “If we can get through this — and I think we will — we’re gonna have a much, much better world.”

Why, in the end, should anyone trust him? Amodei accepts the skepticism as rational and points not to promises but to costs paid: the delay of the first Claude, cutting off model access in China at a price of several hundred million dollars, holding back Mythos. “Look at the overall history,” he asks, “and say, what is the hypothesis about us that is most consistent with that overall history?” His own answer: “We are genuinely trying to do the right thing.” Imperfect, dysfunctional, foot-faulting — but earnest. The interview ends where a man on a relativistic ship might expect it to. “We will see you on the other side of the exponential, then.” His reply: “Hopefully.” And then a confession about the one thing the CEO job never warned him about — how often he has to wear makeup. “That was not on my bingo card.”