Pentagon CTO ‘pretty confident’ about life after Anthropic - Breaking Defense
When Defense Secretary Pete Hegseth gave Anthropic 180 days to comply with the supply chain risk designation, it looked like a countdown to an existential moment for the company. Pentagon CTO Emil Michael suggests the Pentagon sees it differently.
"We have four frontier models. All of them are proving to be largely interchangeable from the user perspective," Michael, the undersecretary for research and engineering, said at the McAleese Defense Programs conference Tuesday. On replacing Claude before the six-month deadline imposed by President Trump, he said he was "pretty confident."
The framing is a blunt instrument: the model Anthropic built its reputation on — Claude, the first frontier system cleared for classified use — is functionally equivalent to three others already in the queue. OpenAI is already deployed. Google's Gemini is next. xAI is in the approval process.
The message, delivered at a defense industry conference in front of contractors and military officials, is also directed at Anthropic: the leverage is not what the company thought it had.
Until February, Anthropic was the only AI company with a model cleared for the Pentagon's classified cloud. Claude ran inside the Maven Smart System alongside Palantir and Amazon Web Services, providing intelligence analysts with a unified picture from multiple sensor feeds. It was used in the operation that captured Venezuelan President Nicolas Maduro in January. Lt. Gen. Jack Shanahan, the former director of the Pentagon's Joint Artificial Intelligence Center, called it "the single most widely deployed AI system in the US military."
That position evaporated in weeks. OpenAI signed an agreement with the Pentagon through Amazon's cloud division in late February, securing the classified work Anthropic vacated. The company — which previously limited itself to unclassified federal contracts — now has the mandate Anthropic once held.
Cameron Stanley, the Pentagon's chief digital and AI officer, told Bloomberg the transition from Anthropic's models currently used in US operations in Iran would take "more than a month," though work is underway. The department is also developing its own LLMs in government-owned environments, with engineering begun and operational deployment expected "very soon." The Pentagon, in other words, is simultaneously plugging the gap with existing vendors and building alternatives for the long run.
Emil Michael said hours after Anthropic filed its lawsuit against the government that the military was moving on and there was little chance of reviving talks. Thomas Kurian, CEO of Google Cloud, had met with Michael on February 26 to pitch Gemini as a replacement.
The interchangeable-models argument has a practical test ahead. Claude had a specific reputation among military users for its interface — analysts found it easier to work with than alternatives. Whether Gemini or OpenAI's models replicate that operational advantage at scale is an open question. Stanley acknowledged the transition will take time. And the Palantir-built Maven Smart System, which Claude runs inside, was not designed with multiple model backends in mind — re-engineering that integration is not trivial.
But the message from the top is clear: Anthropic's safety red lines — no autonomous weapons, no mass surveillance of Americans — are not considered switching costs by the people making the decision. They are reasons to switch.
Sources: