Pentagon CTO ‘pretty confident’ about life after Anthropic

WASHINGTON — The Pentagon’s Chief Technology Officer said he’s “pretty confident” that the military can replace Anthropic’s Claude before President Donald Trump’s six-month deadline, arguing that the four “frontier” AI models were proving largely interchangeable.

“We’ve already deployed OpenAI in the last few weeks, and we’re going to deploy the others here, [starting] with Gemini,” Undersecretary for Research and Engineering Emil Michael said at the McAleese Defense Programs conference Tuesday. “So as these things move up-echelon into different classification networks, we’re going to have tons of different options.”

“What we’re seeing so far is the workflows are very similar,” he emphasized, “so the disruption is, we think, minimal.”

Despite similarities to Anthropic’s Claude, Michael acknowledged that the three other “frontier” Large Language Models — OpenAI, xAI and Google Gemini — each have their own strengths and weaknesses. Claude was “particularly” good at coding, he said, while xAI was good with “real-time content” and Gemini benefited from Google’s control of YouTube.

Michael declined to delve into Anthropic’s ongoing court case against the government, in which the Defense Department filed its official defense [PDF] Tuesday. The Pentagon has officially designated Anthropic as a “supply chain risk” for trying to impose stricter ethical limits on military use of its AI than those the law and regulation already required, which Michael denounced as undemocratic.

At McAleese, Michael framed the issue in terms of avoiding over-reliance on a single supplier.

“One of the problems — without getting into the specifics about Anthropic — is that we had one primary provider,” he said. “That doesn’t work for the Department of War. That doesn’t work for any complex organization. We have to have other options … ideally all options, and then maybe you narrow them over time.”

Such a weeding-out process might take place “over time,” Michael explained, because the models “may all converge” in the future: As their algorithms learn on similar use cases, they might become more and more interchangeable.

Speaking about innovative technology more broadly, the undersecretary emphasized his hope that ongoing acquisition reforms would encourage a rapid growth in “new entrants” to the defense business — without those companies running into undue bureaucratic or legal barriers.

“We hope to have new entrants come, that don’t have to come with an army of lawyers, and [lots] of paperwork, file a lawsuit, hire an auditor and lobbyists,” Michael said. “We’re trying to make it easy.”