Anthropic CEO Dario Amodei Details Pentagon Standoff and Mythos Release Strategy Amid $965 Billion Valuation
Bloomberg interview, June 17, 2026
Anthropic CEO Dario Amodei defended the company's decision to withhold its most powerful cybersecurity model from public release while simultaneously justifying partnerships with the Pentagon that critics say enable AI-assisted military targeting. The apparent contradiction sits at the heart of mounting tensions as the five-year-old startup reaches valuations approaching one trillion dollars, surpassing OpenAI in both market capitalization and revenue growth.
Mythos Withholding Draws Commercial Penalty
Amodei confirmed that Anthropic's decision not to release Mythos, its most advanced cybersecurity model, has "hurt us enormously commercially." The model, which Amodei said found 271 new vulnerabilities in Firefox alone, demonstrated such powerful exploit capabilities that early enterprise customers told Anthropic "this is a super weapon" and requested the company not release it publicly. "You should have to own a gun license to use it," Amodei recounted hearing from multiple customers.
The decision represents a significant financial sacrifice. Mythos has "incredibly accelerated research within Anthropic and production and next models," according to Amodei, and would presumably do the same externally if released. The company is now gradually expanding access to cyber defenders while waiting for stronger safeguards, despite pushback from researchers claiming they could replicate capabilities using cheaper open-source models.
Amodei dismissed those claims as fundamentally misleading. "The idea is Mythos looks across the whole code base and finds something. Some guy went on Twitter and said, well, if you point an open source model at exactly the line of code that Mythos finds, then it finds the same issue. That isn't the prompt, that isn't the question," he said. The company has discovered thousands of vulnerabilities in private companies that cannot yet be disclosed, demonstrating capabilities no previous model achieved.
Pentagon Contracts and Iran School Strike
The interview turned confrontational when Bloomberg reported that Claude is being used by the US military for AI-assisted targeting via Palantir's Maven Smart System, and specifically asked about a February strike on a girls' school in Iran that killed more than 150 people, most of them children. Amodei acknowledged he doesn't know exactly how the models were used in specific operations, saying "we don't have access to" that information.
He defended the partnership framework while acknowledging the tragedy. "Obviously these things that mistakes that happen in warfare are really, really terrible," Amodei said. But he argued the use case doesn't violate Anthropic's red lines because "a human made that final call, not Claude." His concern centers on scenarios where AI makes autonomous decisions without human oversight, which Anthropic's contracts explicitly prohibit.
The position reveals the complexity of Amodei's stance. He described himself as having had a "longstanding anti-war stance" dating to his Caltech days, yet positioned the Pentagon work as essential national security. "When I see Russia invading Ukraine, when I see the risk of China invading Taiwan, it worries me that we have a kind of resurgent authoritarian block that they're very aggressive and that we need to defend ourselves," he said.
Amodei was emphatic that Anthropic doesn't work with ICE or CBP, and doesn't work in Gaza, carefully scoping engagements despite broader Palantir relationships. The company's red lines on mass surveillance and fully autonomous weapons led to the Pentagon recently banning Anthropic from federal contracts and labeling it a supply chain risk, after which OpenAI signed the contract Anthropic rejected.
Revenue Growth Dramatically Exceeds Infrastructure Planning
Amodei revealed striking numbers on Anthropic's recent growth trajectory that explain recent reports of server strain and token rationing. The company saw "greater than three times growth in revenue quarterly" in the first quarter of 2026, which annualizes to approximately 80 times growth. "We didn't plan for 80x annualized growth. It would not have been rational to plan for 80x annualized growth," he said.
The company had been planning for 10x annual compute growth, which Amodei characterized as reasonable. But the actual growth rate created what he called a "locally extreme explosion of compute" that cannot and will not continue at that pace, since it would lead to revenue numbers exceeding any company on earth. The mismatch has created short-term capacity issues despite large compute deals with Google and Amazon.
On reports of approaching trillion-dollar valuations, Amodei was characteristically analytical. "I was watching this graph for a while and 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," he said. The smooth exponential he's been tracking for years creates the paradox of being simultaneously unsurprised by the trend line while surprised by the detail and color when events actually materialize.
Enterprise Focus Drives Model Leadership
Amodei explained the strategic decision to focus on enterprise rather than consumer applications as both a values and business model choice. "If you pick a business model that fundamentally conflicts with your values, you're gonna have a hard time," he said. "Either you betray your own values or you become irrelevant."
The consumer social media world encourages "engagement, even addiction" and what Amodei called "slop" from AI video models, driven by advertising revenue incentives tied to attention minutes. Enterprise, by contrast, aligns with positive use cases Amodei wants to enable: curing diseases through biotech partnerships, making energy cheaper, improving education, and increasing economic growth.
Crucially, enterprises "care a lot about trust in long-term relationship" rather than the "gimmicky aspect" that can characterize consumer products. This synergy with Anthropic's safety-focused mission reduces conflicts between commercial success and stated values, though Amodei acknowledged hard choices still arise.
The strategy is paying off. Even consumer usage is accelerating "without us putting that much effort," Amodei noted, while Claude Code and Claude Cowork have become hits. The February release of Claude Cowork triggered what traders called the "SaaSpocalypse," erasing $285 billion in traditional software market value overnight.
Job Displacement Warnings Spark Backlash
Amodei pushed back hard against critics like Jensen Huang who characterized his warnings about AI-driven job loss as "doom marketing." He noted that he's consistently paired warnings with detailed solutions across multiple essays, including discussions of token taxes, enterprise partnerships for worker adjustment, macroeconomic policy, and the distinction between task automation and job elimination.
"The whole picture of there are risks to job loss and here are some ideas, I mean, we haven't fully fleshed out the ideas because I want to get them right, but Anthropic has come up with lots of ideas," he said. His frustration centered on social media's tendency to extract "three second clips" that ignore the nuanced analysis in his longer writings.
"This is laziness, this is failure to engage with serious intellectual work," Amodei said of the criticism. "I think it's part of the disease of Silicon Valley. It's been caught up in this social media world of three seconds. Whenever someone says something like that, I take them less seriously."
He maintained his order-of-magnitude estimate that AI could eliminate roughly half of entry-level white collar jobs over one to five years, though he emphasized massive uncertainty around timing and extent. Within Anthropic itself, the company is already seeing the beginning of software engineers whose productivity isn't increasing from AI assistance, where "it's better for the AI to just do the thing."
The company is actively pushing enterprise customers toward expansion rather than cost-cutting uses of AI. "They face the choice of should I save costs, which often means hiring less people, basically do the same thing with less resources, or should we do more things with the same amount of resources," Amodei said. "We always, when we can try to push them to doing more with the same amount of resources."
OpenAI Departure Rooted in Trust Breakdown
Amodei offered his most direct comments yet on why he and other Anthropic co-founders left OpenAI, moving beyond previous diplomatic statements. "There are many valid disagreements to be had on safety. We certainly had some of those disagreements with them, but people here have had disagreements with me, people here have disagreements with each other," he said.
The actual breaking point was 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 feel that they're not in it for the reasons that they say, when you see disturbing patterns of behavior, dishonesty, that makes it very hard to continue to work with a company, to continue to trust the company."
He framed the separation as the natural resolution rather than ongoing conflict. "Why argue with someone when you don't have the same vision and you don't trust them. The way to resolve it is you go off and do your thing. They go off and do their thing," he said. "We'll see who wins in the market and we'll see who wins in the court of public opinion."
On the viral moment at India's AI summit where he appeared to refuse holding hands with Sam Altman on stage, Amodei blamed disorganization. "All of these kind of international type summits that have heads of state are super disorganized," he said, noting they changed the standing order at the last minute before suddenly ordering everyone to hold hands.
China Strategy and Export Control Support
Amodei disclosed that Anthropic voluntarily cut off access to Chinese users despite no legal requirement, a decision that "cost us several hundred million dollars back when several hundred million dollars was a significant fraction of our revenue." He's been publicly advocating for export controls on AI chips to China, a position that conflicts with some of Anthropic's chip supplier partners.
"I say this because I think it would be really bad for America, for the state of democracy in the world for China to be ahead in AI capabilities," he said. "Some of the chip makers obviously don't agree with that view, but it hasn't stopped me from saying it. I'm saying it again now, even after we've signed more partnerships."
His year at Baidu early in his career didn't significantly shape these views, though he noted being told "ominously" that "we don't care about privacy in China" when discussing how they obtained speech recognition data. His concerns center on what he's observed regarding Uyghur suppression, Hong Kong, and the CCP's ability to "reach into the US business network and suppress criticism."
"That's an authoritarian state and a high tech authoritarian state. And when I see how that combines with AI you really get a dystopia here like 1984 or worse," Amodei said. He sees an opportunity for AI to be "a pro-democracy technology" that delivers on equal justice promises, but warned "it could go the other way."
Self-Improving AI Already Accelerating Development
Amodei confirmed that AI is already beginning to improve itself, though he rejected the notion of a discrete "moment" when this begins. "We're already seeing it in some ways where the AI is able to suggest architectures for the next AI," he said. A year ago, AI contributed 10 to 15 percent increases in total factor productivity for model development. "That's probably up to 20 or 30% now, might be doubling."
This acceleration is visible in Anthropic's product velocity, which Amodei attributed to two factors: unified company culture that maintains efficiency despite rapid growth, and "Claude itself, that we're now using Claude to help develop our models and make them more efficient and quickly develop products."
He emphasized that recursive self-improvement is "a continuous process" rather than a dramatic threshold. "There's no moment where AI improves itself or runs out of control or becomes unsafe. What we have is an accelerating exponential and at each point on the exponential we have to assess is this a time to slow down, is this a time to put more controls on this technology."
Civilizational Risk and Accountability Structures
Amodei maintained his estimate of roughly 10 to 25 percent probability of civilizational collapse from AI, while arguing Anthropic's actions lower rather than raise that probability. "That probability comes from the very straightforward recipe of the technology, the existence of many countries in the world, the existence of many companies within an economy and new ones created," he said.
He drew parallels to nuclear weapons and the internet, noting that AI is "the first technology that's been built in the private sector and where government has not really had a serious role and is coming in late to the game." He called this "a dangerous and unstable situation" though not one he would have chosen, since "this technology is possible to build, our adversaries are building it, it has economic value."
Anthropic's Long Term Benefit Trust structure is designed to provide checks on company power. The trust can appoint and remove the majority of board members, which "essentially, if you thread it through, has the power to fire me," Amodei said. "We're introducing a little bit of the elements of public governance where you're accountable to someone who doesn't just have stock in the company."
He called for reciprocal checks, with companies providing oversight on government and government providing oversight on companies. "I'm scared of companies having it, but I'm also scared of government having it," he said. He criticized the tendency to yo-yo between extreme anti-regulation positions and calls for complete nationalization, advocating instead for "sensible, moderate" approaches including required pre-release testing and auditing.
Relationship Dynamics with Google and Amazon
When asked how Anthropic maintains independence given massive backing from companies with their own AI agendas, Amodei pointed to policy disagreements he's aired publicly despite partnerships. His advocacy for chip export controls to China is one example where "some of the chip makers obviously don't agree with that view, but it hasn't stopped me from saying it."
"I'm sure they wish we didn't say these things, but these things are what I believe. What are you gonna do?" he said. "They're at the end of the day, they want, they benefit from these deals as much as we do. We're all adults here. We can work together on one thing while disagreeing about another thing."
He noted productive collaboration with Google's DeepMind, led by Demis Hassabis whom Amodei has known for 15 years. "We buy compute from Google, we swap safety ideas all the time," he said. This represents the positive "carrot side of the race to the top" where companies inspire each other through innovations like AlphaFold in biology or interpretability research.
Trust Must Be Earned Through Actions
When confronted with the fundamental question of why the public should trust a company building enormously powerful technology from which it stands to gain tremendously, Amodei acknowledged starting from a position of distrust is rational. "Silicon Valley has lost a lot of the world's trust and kind of has to re-earn it," he said.
He pointed to a track record of costly decisions aligned with stated values: the Mythos release delay, cutting off Chinese access worth hundreds of millions in revenue, the Claude 2 delay, and the Pentagon standoff. "We aren't perfect, we make mistakes. But what I would ask is for people to look at the overall history and say, if you add up that overall history, what is the hypothesis about us that is most consistent with that overall history."
The hypothesis he offered: "We are genuinely trying to do the right thing. We're imperfect, organizations are always dysfunctional, we're always trying to fix them and make them work better. Many foot faults, many things that go wrong, but at basis we have an honest and earnest picture of how to do the right thing and we're trying to execute on that picture."
Amodei rejected comparisons to Oppenheimer, saying he most identified with Leo Szilard who first conceived of nuclear chain reactions. "My view is we're not gonna get through this with larger than life personalities or figures who try and be at the center of everything," he said. "There needs to be a balance of power here. In some ways I actually see Oppenheimer as a failure case, as what should not happen."