The Pentagon is on track to complete its migration of all classified AI workloads away from Anthropic's models by the end of October, according to Under Secretary of Defense for Research and Engineering Emil Michael, who told reporters that roughly 90% of the transition is already done.
"I'd say about 90% has transitioned. All of the Maven Smart Systems or Palantir work has been transitioned months ago, and we're on track to get it done by the end of the month," Michael said at the National Defense Industrial Association's Emerging Technologies for Defense Conference, as reported by DefenseScoop on September 11. For more context on this story, see our ongoing AI trends.
How the Break With Anthropic Happened
The fast-tracked migration is the result of a very public collapse in negotiations between the Defense Department and Anthropic, one of the leading U.S. frontier AI labs. According to DefenseScoop's reporting, Anthropic insisted on contract safeguards that would have restricted its models from being used for certain applications — specifically, mass surveillance of U.S. citizens and fully autonomous lethal weapons systems.
The department rejected those terms, arguing that defense software must be available for "all lawful purposes" in an unconstrained way. The standoff escalated: the Pentagon subsequently designated Anthropic as a national security supply chain risk, and the two parties are now fighting it out in court.
Michael offered new detail on the litigation landscape. He said Anthropic was designated as a supply chain risk under two different laws, with one case in the Northern District of California already adjudicated — Anthropic, he said, "forum-shopped" that venue. The more consequential case, in his view, sits with the D.C. Circuit Court, which he described as a more national-security-oriented forum.
"We should hear from them soon, [in] the next month or two," Michael said, noting that Anthropic remains a supply chain risk in the department's eyes because the court has not granted a preliminary injunction in its favor.
The Multi-Model Pivot
Beyond the dispute itself, Michael framed the episode as a lesson in procurement strategy. "One of the mistakes, independent of whether we had a dispute with Anthropic, is the fact that the Pentagon was solely reliant on one model," he said at a media roundtable. "What if that model were [Google's] Gemini or some other model that hadn't kept up with the frontier? We would be putting our warfighters in a disposition that was suboptimal."
The department's answer is a multi-model ecosystem that deliberately spreads generative AI capability across several frontier vendors rather than concentrating it with one. That ecosystem spans both classified networks and unclassified operational platforms such as the nascent GenAI.mil program, which has been training service members to build AI agents — including a July deployment to U.S. Pacific Fleet sailors at Joint Base Pearl Harbor-Hickam in Hawaii.
Technology suppliers, deployment methods and security levels all differ depending on the workspace, meaning the Anthropic exit touches intelligence and warfighting systems of very different sensitivity. The Maven Smart System, the Pentagon's flagship AI-powered targeting and analysis platform, and its Palantir integrations were moved off Anthropic months ago, per Michael's account.
Early Deployments: Enthusiasm Meets Reality
Officials speaking on condition of anonymity offered DefenseScoop a mixed picture of the early enterprise deployments of military-tuned frontier models. One said the tools are already woven into daily work — and that leadership is mandating adoption. "I'll use it for anything I can," the official said. "It is not perfect and it's not the end all be all, but I've now actually given my team an edict — I said, 'If you're going to bring me a product and you haven't run it through AI at least once, you don't get to come into my office.' And part of it is to force that learning, because it's a different skill."
Another official urged the department to "not oversell where we are in the AI journey," offering a striking personal comparison: they pay for Anthropic's Claude out of pocket for use at home, because in 20 minutes they can complete work that takes four hours on government systems. The gap, they explained, is architectural — GenAI.mil currently lacks API access, cannot run custom agents and is not integrated into tools like email, so the "hyper-productivity" available outside the "dot mil" domain does not yet exist inside it. "We're not there yet. I think there's a vision to get there, but we have a lot of work to do," the official said.
What It Means for the AI Industry
The dispute has become a defining test case for how frontier AI labs and national security customers do business. Anthropic's position — that its models should carry usage restrictions even for government customers — reflects the safety commitments the company has built its brand around. The Pentagon's counterposition — that anything lawful must be on the table — reflects a decades-old defense procurement philosophy that commercial vendors serve all lawful military needs.
Other labs have quietly positioned themselves to fill the gap. With roughly 90% of classified workloads already transitioned, the practical consequences of Anthropic's stance are now measurable: one of America's leading AI companies has been largely excluded from the classified systems where frontier AI is being operationalized, from intelligence analysis to targeting support.
The coming weeks will show whether the migration closes on schedule and whether the D.C. Circuit's ruling reshapes the standoff. Either way, the message to the AI industry is now on the record from one of the Pentagon's top technology officials: restrictive licensing terms carry a real cost, and no single model is indispensable.
Stay Ahead of AIGovernment AI policy is shifting week to week. For breaking AI news, sharp analysis of AI regulation and the latest AI developments across the industry, follow AI Buzz Wire.
Read more AI news
