Emil Michael, the under secretary of Defense for research and engineering, said Thursday that the Pentagon will not reconcile with Anthropic, despite softening rhetoric toward the AI company from the White House.
Speaking at the AI+ Expo in Washington, Michael told The New York Times' chief Washington correspondent David Sanger that the Department of Defense had learned its lesson after becoming dependent on a single AI vendor. "Never again will we be single-threaded with any one model," Michael said. "We were singled-threaded with Anthropic."
The Pentagon blacklisted Anthropic approximately two months ago following a dispute over potential uses of its AI models for domestic surveillance or fully autonomous weapons systems. Defense Secretary Pete Hegseth subsequently banned the company from military work, and President Trump extended the restriction to civilian agencies.
When Sanger asked whether Anthropic could eventually sign deals to deploy its models on classified networks, Michael responded: "not at the Department of War."
Michael noted the technical complexity of integrating AI vendors into classified systems, explaining that these are sophisticated protective environments requiring extensive work. "To integrate into a classified system is not just putting your software on a public cloud and having it work," he said. "So it wasn't like I could just turn on a few other models that easily."
Eight companies including Microsoft, Google and xAI signed similar AI deployment deals with the Pentagon last week.
What the Right Is Saying
Conservative defense hawks have supported the Pentagon's stance, arguing that companies unwilling to guarantee their technology will not be used for surveillance or fully autonomous combat should not have access to classified military networks. They say dual-use AI capabilities require strict vetting.
Republican officials point to Trump's own comments as evidence of calculated diplomacy rather than policy reversal. The president described Anthropic's team as "very smart" with "high IQs" but emphasized building a diverse vendor base for national security.
Conservative commentators have framed Mythos's release as a negotiating tactic, arguing that the company's public demonstration of cybersecurity capabilities is designed to pressure the administration into lifting restrictions rather than representing a genuine offer of cooperation on defense terms.
What the Left Is Saying
Progressive critics of the Anthropic ban have argued that excluding the company limits the Defense Department's access to leading-edge AI capabilities during an intensifying technological competition with China. They contend that cybersecurity threats require partnerships with the most advanced developers, regardless of policy disagreements.
Democratic lawmakers have questioned whether the administration's broad prohibition on civilian agency use of Anthropic products amounts to overreach, noting that companies like Anthropic employ safeguards against misuse that competitors may lack.
Civil liberties advocates aligned with Democratic priorities have praised the ban on autonomous weapons applications while arguing that Mythos's cybersecurity capabilities represent exactly the kind of defensive AI tools government should be deploying rather than restricting.
What the Numbers Show
Eight AI companies now hold Pentagon contracts for classified network deployment, compared to one—Anthropic—before the ban. Anthropic's Mythos model reportedly can detect vulnerabilities spanning decades across web browsers, infrastructure and software systems. Treasury Secretary Scott Bessent, Federal Reserve Chair Jerome Powell and White House chief of staff Susie Wiles have each held separate discussions with Anthropic representatives about Mythos since its release, according to administration officials. The Pentagon has not disclosed the timeline or cost of transitioning away from its previous Anthropic integration.
The Bottom Line
The split between Pentagon and White House approaches to Anthropic reflects a broader tension in AI policy: whether cybersecurity imperatives require engaging all capable providers or whether national security concerns justify excluding companies that refuse commitments on autonomous weapons use. Michael's framing of Mythos as a "cyber moment" suggests the debate will expand beyond this single company to encompass how the U.S. government operationalizes vulnerability detection and defense at scale. Watch for any congressional action on AI procurement authority, which could formalize or limit executive discretion over vendor selection for sensitive systems.