Pentagon CTO Says Anthropic Won’t Be DOD Vendor, but Mythos Raises ‘National Security’ Stakes

The Pentagon will not use Anthropic as a Department of Defense (DOD) vendor despite ongoing government interest in the company’s powerful Claude Mythos artificial intelligence (AI) tool, DOD Chief Technology Officer (CTO) Emil Michael said Tuesday at the Politico Security Summit.

Michael described Mythos, which includes powerful hacking capabilities, as a “national security moment” that could reshape how the federal government approaches AI-driven cyber defense.

His remarks offered new insight into how the Trump administration is attempting to separate Anthropic’s advanced cyber models from the broader dispute that led the company to be designated a federal “supply chain risk” earlier this year. That designation followed Anthropic’s refusal to remove certain guardrails on its AI technology for military use.

The Trump administration has rebranded DOD to the Department of War.

“They’re not going to be a vendor to the Department of War,” Michael said, referring to Anthropic. “We have six months to get off of Anthropic.”

At Tuesday’s summit, Michael said the Pentagon cited recent agreements made with SpaceX, OpenAI, Google, Nvidia, Reflection, Microsoft, Amazon Web Services, and Oracle to deploy advanced AI capabilities across its classified networks.

“The whole industry agrees that they have to serve the Department of War in the same way, except for one,” Michael said, referencing Anthropic.

Still, Michael acknowledged that Anthropic’s Mythos model presents a separate and more urgent challenge for government officials. He described Mythos as part of an emerging class of AI cyber models that could dramatically accelerate both vulnerability discovery and cyber exploitation capabilities.

“We’ve never seen this before,” Michael said. “When you have a model that’s capable of doing really exquisite vulnerability finding and then exploitation, every part of the government and the American industry … has to be concerned.”

“The government, writ large, not just the Department of War, has to look at that capability,” Michael added.

A ‘national security moment’

Last week, OpenAI announced a new tool – labeled GPT-5.5-Cyber – that has been touted as a competitor to Mythos. Michael predicted that competing models from Google, xAI, and other AI companies will soon follow.

“[Mythos] is the first of many,” he said at the summit. “It’s being treated as a national security moment, so to speak, and has to be used to harden up systems across the government and across the country.”

Despite the Pentagon’s hard line against Anthropic as a vendor, Michael indicated the administration sees frontier cyber AI as too strategically important to ignore.

“The use case for Mythos across the government is to find and patch your vulnerabilities, and we’ll be doing the same thing for OpenAI,” he said, adding, “Those things are sort of a different category … It’s using them to figure out what they can do. You know, what is the response that the government has to have? What defenders are we bringing in to defend the networks?”

Rep. Jim Himes, D-Conn., ranking member of the House Intelligence Committee, said at the same event that Mythos is forcing lawmakers and national security officials to more urgently confront the cybersecurity risks posed by frontier AI systems.

“Mythos is raising a very precise question about unimagined hacking capabilities in the hands of our adversaries,” Himes said, arguing that the technology presents an “act now problem” for both Congress and federal agencies.

Himes also cautioned that prolonged conflict between the Trump administration and Anthropic could ultimately become “a massive liability for United States national security.”

“We ought to be cultivating, not damaging, our relationship with the producer of this remarkable new technology,” Himes said.