Anthropic CEO Dario Amodei says he left OpenAI because he “don’t trust them”

Anthropic CEO Dario Amodei didn’t dodge the real reason he left OpenAI. In a Bloomberg interview for “The Circuit with Emily Chang” published Wednesday, he framed his departure from OpenAI and Sam Altman around a blunt idea: “Why argue with someone when you don't have the same vision and you don't trust them?” That line is more than a personal confession. It is a thesis about how AI companies should coordinate, compete, and decide what “trust” even means when the stakes are existential.

Amodei also said he is at “peace” with the rivalry between Anthropic and OpenAI, even as it hardens into what the article describes as a cold war between two of the world’s largest AI companies. He argued the resolution is not debate club or hand-wringing. “The way to resolve it is you go off and do your thing, they go off and do their thing,” he said. “And I am completely at peace with the idea that we're doing things our way and they're doing things their way.” Then he pointed to the scoreboard he believes matters: “We'll see who wins in the market and we'll see who wins in the court of public opinion.”

To understand why this moment matters for leaders, you have to zoom out to the “trust” problem now haunting AI. According to the source, Amodei’s remarks got renewed attention amid Ronan Farrow’s exposé in The New Yorker, which examined whether Altman could be trusted. The report cites Amodei’s contemporaneous notes about his interactions with Altman during his time at OpenAI, and those notes have been pulled into the public conversation as evidence in a story about credibility. Days after the New Yorker report’s publication, Altman’s home was attacked, an incident he partially blamed on The New Yorker report (without naming the publication directly).

This is the context where Amodei’s “trust” line becomes strategically important. When AI labs disagree on methods or safety priorities, the disagreement often gets reframed as moral failure, not just engineering divergence. The article also says Amodei responded indirectly to how the narrative has shifted. During an April episode of the “Core Memory” podcast, Altman said, “I think the doomerism talk hasn't helped,” and he added that “the way certain other labs talk about us hasn't helped.” He also criticized “the way Anthropic talks about OpenAI.” In other words, both sides are describing the same phenomenon from opposite ends: rhetoric changes trust, trust changes cooperation, and lack of cooperation changes what gets built.

Amodei gave his own counter-narrative. He pushed back on the idea that “no one trusts each other” in AI, citing his relationship with Google DeepMind CEO Demis Hassabis. “I've known him for 15 years,” Amodei said. “We've worked together on a number of issues.” He added concrete examples: “We buy compute from Google. We swap safety ideas all the time.” The point is practical. He is not claiming universal trust across every lab. He is arguing that some actors are trustworthy enough to coordinate with, and that labs can decide who sets the agenda.

That brings us to the policy and safety dimension, where these personal trust dynamics translate into industry coordination mechanisms. When Chang asked how the world could trust AI companies to cooperate on major AI safety issues, Amodei suggested it is “less about everyone getting along and more about who sets the agenda.” He said, “What I think needs to happen is that the trustworthy actors need to get together and put the untrustworthy actors in a position where they kind of have to adopt the same standards.” Then he added a hardball version of what many safety advocates would call “market discipline,” but he framed it as learned experience: there are some actors who “don't do the right thing on their own,” but if a majority is doing it, the rest face pressure and “they're left in a position where there's not much they can do.”

This isn’t just talk. The article says Amodei’s comments about rivalry have been building around public moments that the internet treated like signals. One viral episode involved Amodei and Altman declining to join their fellow industry leaders in a show of unity at an India AI summit. Amodei attributed the viral moment to the “extreme disorganization” of the summit. He also described other summits as “often a mess as well,” including a remark that Prime Minister Narendra Modi appeared and “told everyone to hold hands.” The source notes that Modi was positioned right next to Altman and Amodei, and that left unsaid is that both held hands with others next to them, while “Sam Altman and Dario Amodei's hands did not make contact.” In other words, the public read body language as governance.

The second-order implication for executives is that safety coordination in AI is now inseparable from reputational conflict. If trust is personalized, then cooperation becomes fragile, and the “court of public opinion” Amodei mentioned becomes part of the compliance environment. Board members and leaders at competing labs will likely hear his message as: do not assume goodwill will carry you through. Instead, build safety standards with actors you believe are trustworthy, and let the market and reputational dynamics do the enforcement when consensus fails.

The other implication: when companies treat rivalry as separable strategies rather than a shared mission, regulators and governments may face a more complicated landscape. The source does not claim new regulatory actions, but it does highlight how public narratives, major media scrutiny, and even personal security events around founders are shaping the trust atmosphere. For any CEO or investor, the lesson is uncomfortable: in this market, “trust” can become a product feature, a competitive moat, and a political accelerant all at once.