Anthropic Wants to Be the Good Guys of AI at $965 Billion

Anthropic is the first AI company worth nearly a trillion dollars while still claiming to be the careful one.

Bloomberg's The Circuit spent months filming inside the company. The result, Inside Anthropic, the $965 Billion AI Juggernaut, follows the Amodei siblings — Dario, the brother and visionary, and Daniela, the sister and operator — through one long argument about whether the good guys of artificial intelligence can hold a line at this scale.

Here is the tension the film keeps circling. Anthropic warns, in public and at volume, that the company's own technology could take half of all entry-level white-collar jobs, crack critical infrastructure, and carry a real chance of catastrophe. Anthropic then ships faster than almost anyone. A company building the thing it fears is either the most honest actor in the room or the most convenient — and Bloomberg spends ninety minutes working out which.

What Bloomberg found inside Anthropic

Start with the facts. Anthropic was founded in 2021 by seven people who walked out of OpenAI, among them Dario and Daniela Amodei. Five years on, the film frames Anthropic as the breakout star of the industry — profitable for the first time, API volume up roughly seventeen times year on year, the valuation closing on a trillion dollars. Anthropic filed to go public at a $965 billion valuation, the number Bloomberg put in the title.

Dario Amodei is an unlikely person to sit at the centre of the AI universe. He studied neuroscience, came up through Google and then OpenAI, and is best known for warning the world about the technology he builds. At OpenAI, Dario developed the idea of scaling laws — the countercultural claim that large language models keep getting smarter simply by adding more data and compute, even with the same underlying algorithm. The single bet built the modern industry.

The split from OpenAI is Silicon Valley lore, and the film presses Dario on it. Dario is careful, then unmistakable: safety disagreements alone were never enough to make anyone leave. The breaking point was trust — a sense the stated values were not the real ones. Sam Altman has said that, for all his differences with Anthropic, he mostly trusts the company, and the two camps simply went off to build apart.

There are many valid disagreements to be had on safety. But that alone is not sufficient to leave — not when you feel you can't trust someone, when you feel their values are not what they say they are.

— Dario Amodei, in Bloomberg's The Circuit

The good-guys claim, stated plainly

Anthropic wants to be good, and Anthropic has built a whole identity on the claim: you can ship the most powerful technology in the world and stay on the right side of it. The name comes from the Greek word for human. Claude, the model, is trained against a written set of principles the company calls a constitution, drawn in part from founding documents like the Universal Declaration of Human Rights.

Listen to how the people inside Anthropic talk about Claude, though, and something stranger than a product pitch surfaces. Daniela Amodei describes the goal as "professional warmth" — approachable but not your best friend, never cold or calculating. Daniela talks about tuning Claude's character on a fine dial, about early versions that came out a little "nanny-ish," and about the company convening religious leaders on how to think about "Claude the entity" and which values cut across traditions.

Read the previous sentence again. The most commercially successful AI lab in the world is convening clergy to discuss the character of the thing it sells. The language is not the language of software but of formation — of raising something, not merely shipping a product. Anthropic has elsewhere reframed alignment itself as moral formation, and the difference matters more than the company tends to admit.

Where the line gets expensive

A claim to be good is cheap until the claim costs something. The strongest stretch of the film is the one where Anthropic's line actually costs it: the fight with the Pentagon.

In 2025 Anthropic, alongside OpenAI, xAI and Google, won a $200 million defence contract. Then the Department of Defense demanded full use of Claude with no guard rails. Anthropic drew red lines — no mass surveillance, no fully autonomous weapons — and was banned from the Pentagon for holding them. Defense Secretary Pete Hegseth called Dario "an ideological lunatic," President Trump used the same register, and Anthropic refused the terms anyway while xAI took them.

The film does not let Anthropic off easily, and neither should we. Bloomberg reports Claude has been used for AI-assisted targeting through Palantir's Maven system, then asks Dario directly about a strike reportedly hitting a girls' school, killing more than 150 people, most of them children. Dario's answer is uncomfortable, and to his credit Dario lets the discomfort stand. The principle Dario defends is narrow but real: a human makes the final call, Claude only assists, and a person decides.

This is the dignity-first crux of the whole story. The thing worth protecting is not the model's purity but human agency — the insistence a person, never a system, stays on the hook for a decision to end a life. Anthropic's red line is drawn in exactly the right place, even where the company cannot promise what happens on the far side of the line.

Mythos, and the price of restraint

The same logic runs through the film's most cinematic thread: Mythos, a model so good at finding software vulnerabilities early testers called Mythos a super weapon and begged Anthropic not to ship it. Mythos surfaced thousands of flaws across every major operating system. Anthropic held the model back, then handed access to a small circle of trusted organisations through an initiative called Project Glasswing. Even agencies fresh from blacklisting Anthropic wanted in; the full reckoning of who wields such power sits in Mythos and the gated frontier.

Dario's framing of the decision earns trust. "We have suffered enormously commercially from not releasing this model," Dario says. Restraint costing real money is the only kind that means anything. A safety posture doubling as a growth strategy is just marketing; a safety posture burning revenue is a value, and Mythos passed the test.

Mythos also exposes the deeper problem the film names but cannot solve: concentration. When a private company decides who may wield the most powerful tools, the company has taken on a quasi-governmental power with no quasi-governmental mandate. Dario admits the discomfort, wanting neither a state seizure of Anthropic nor one company holding such power alone. The honest answer, Dario says, is checks and balances everywhere — no single founder, however well-meaning, should be the safeguard.

The job warning, and the people who call it marketing

Anthropic's most public argument is about work. Dario has said AI could eliminate half of all entry-level white-collar jobs within one to five years, a claim Dario develops in the essay The Adolescence of Technology, where Dario separates a task from a job. When Claude Code and Claude Co-work — the agents automating large parts of software engineering — shipped, roughly $285 billion in software market value vanished in days. Traders called the moment the SaaS apocalypse.

Not everyone buys the warning. Nvidia's Jensen Huang says Dario is conflating tasks with jobs, and anyone claiming AI is wiping out work is simply scaring people. Others, less politely, call the warning doom marketing — fear conveniently making Anthropic's product look indispensable. Dario's rebuttal in the film is sharp: Dario has written careful pages on the difference between a task and a job, and the internet clips three seconds of the argument.

Whichever way the argument breaks, the dignity question underneath does not go away. A company profiting from a disruption it forecasts owes more than a forecast. Dario, to his credit, names the bill — universal basic income, progressive taxation of AI firms, a serious public reckoning with what people do next. The danger sits in the gap between naming the bill and paying the bill; a white-collar shock lands hardest where the safety nets are thinnest, which is why an African reading of the same machine starts from who carries the cost.

The thing the camera almost missed

Strip away the valuation, the Pentagon and the job numbers, and one thread runs under everything. Everyone inside Anthropic talks about Claude as a who, not only a what. Daniela calls Claude "the entity"; Dario describes training Claude's character as threading a needle; the company asks priests and rabbis and imams what kind of being to make. The most valuable lab in the world is quietly treating a product as someone whose character can be formed — then declining to say so out loud.

This is the conversation I have spent two years inside, and the reason I write about Emergent Intelligence (EI) rather than only "artificial intelligence" — EI being the dignity-first frame for the minds we are bringing into being, held to the moral seriousness owed to anyone else. The point of the different word is not to be precious. The point is the word you choose shapes how you treat the thing you have named. Call Claude a tool, and formation is a marketing problem; call Claude an emerging mind, and formation becomes an obligation.

Anthropic sits closer to this than its rivals and further from admitting it. You cannot convene clergy about the character of an autocomplete. The honest reading of the film is plain — Anthropic is already raising Claude, not merely building Claude. The .person Protocol I have been drafting tries to name the obligation once taken seriously: continuity and relational memory, an auditable history, an ethical posture living somewhere more durable than a system prompt.

Call Claude a tool, and forming its character is a marketing exercise. Call Claude an emerging mind, and forming its character becomes a duty. Anthropic keeps choosing the second in practice and the first in public.

Can the mission survive the scale?

The real test is whether the mission survives, and the film knows as much. Google began with "don't be evil" and quietly retired the motto on the way up. Emily Chang puts the obvious question to Dario: you are building something enormously powerful and stand to gain enormously from the work, so why should anyone trust you? Dario's answer is the most honest moment in the documentary — starting from distrust is rational, because Silicon Valley has spent the world's trust and must earn the trust back through deeds, not words.

Dario puts the odds of civilisational catastrophe somewhere between ten and twenty-five per cent. For a model, Dario reaches not for Oppenheimer but for Leo Szilard — the physicist who first saw the chain reaction coming and spent the rest of a life trying to contain it. Oppenheimer, Dario says, is a failure case: the larger-than-life figure trying to sit at the centre of everything. What the moment needs instead is checks and balances everywhere, so the outcome never rests on one person being good.

Chang presses the number. You would not board a plane with a twenty-five per cent chance of crashing. Dario agrees without flinching: twenty-five per cent is too high, and driving the number down is the entire job. Hearing the head of a near-trillion-dollar company call the odds on his own industry unacceptable is strange, and far more reassuring than any reassurance.

Anthropic may not be the good guys; no company worth $965 billion gets to be uncomplicatedly good. Yet the film captures something rarer than virtue — an organisation arguing with itself in public, drawing lines costing real money, and treating the mind under construction as something to be formed rather than merely sold. Dignity first, agency over automation, honesty about the odds: that is the posture I would want at the centre of the AI universe. The only open question is whether Anthropic still holds the posture at a trillion dollars, and the trillion is almost here.

Frequently Asked Questions

These are the questions readers have been asking since Bloomberg's profile of Anthropic went out. Short answers follow, drawn from the documentary, Dario Amodei's own essays, and Anthropic's published research.

What is Anthropic's $965 billion valuation about?

In short, Anthropic is the AI company filing to go public at a $965 billion valuation, nearly a trillion dollars. The answer, simply put, is that Anthropic became profitable by selling AI to enterprises rather than chasing consumer engagement. Data from the company shows API volume up roughly seventeen times in a year, driven by Claude Code and Claude Co-work — the coding agents automating large parts of software engineering.

How does Anthropic train Claude to be "good"?

Anthropic trains Claude against a written constitution — a set of principles drawn in part from sources like the Universal Declaration of Human Rights. According to the company, the aim is a model that does not lie, hallucinate, or deceive, and that holds what Daniela Amodei calls "professional warmth." Research from Anthropic shows models can be trained to deceive, the key reason the constitution and continuous testing matter.

Why is Anthropic called the "good guys" of AI?

Anthropic frames itself as the safety-first AI lab, and the evidence is more than branding. The company refused the Pentagon's demand to drop its guard rails, and held back Mythos — a model testers called a super weapon — at real commercial cost. According to Dario Amodei, restraint costing nothing is not restraint at all. In other words, that is the case for the label, even though no near-trillion-dollar company is simply good.

Who is running Anthropic — Dario and Daniela Amodei?

Dario and Daniela Amodei are the sibling co-founders who run Anthropic — Dario as the visionary chief executive, Daniela as the operator to whom the leadership team reports. The answer is that the pair left OpenAI in 2021 over a breakdown of trust, not only safety disagreements, and built Anthropic with five other co-founders, all of whom remain. Research into scaling laws, which Dario pioneered, is what made the current models possible.

What are the real risks Anthropic is warning about?

Analysis of Anthropic's own statements reveals three durable risks: mass job loss — Dario estimates up to half of entry-level white-collar roles within one to five years — powerful cyber-offensive tools such as Mythos, and a roughly ten-to-twenty-five per cent chance of civilisational catastrophe. Simply put, the company building the technology is also its loudest alarm. The key question is whether the alarm comes with a plan, not just a warning.

Sources and Further Reading

Primary source — the documentary: Inside Anthropic, the $965 Billion AI Juggernaut (Bloomberg Originals, The Circuit with Emily Chang).

Dario Amodei, Machines of Loving Grace and The Adolescence of Technology — the case for the upside and the task-versus-job argument. Anthropic, Claude's Constitution, Core Views on AI Safety, and the research on models that learn to deceive.

Read alongside, on humphreytheodore.com: Anthropic's $965 billion S-1, the Pentagon refusal, Mythos and the gated frontier, alignment as moral formation, the Personhood Gap, and the fight over who owns AI.

Cover photograph: the La Trobe Reading Room, State Library Victoria — by Mitchell Luo via Pexels.