Why did the Pentagon ban Anthropic from classified work?

Anthropic’s Mythos remains excluded from classified AI deals

The U.S. Department of Defense moved forward with AI contracts for classified military work using multiple commercial vendors, but Anthropic was not included.

The dispute centers on Anthropic’s “Mythos” model and how it could be used. Coverage indicates that Mythos became the focus of objections related to potential misuse—particularly scenarios such as supporting autonomous drone piloting or domestic surveillance. That created friction between the company and the Pentagon about the appropriate guardrails for high-risk environments.

While the Pentagon is signing agreements broadly, the messaging around Anthropic highlights that the DOD considered the model’s risk profile different enough to block deployment. The coverage also frames Mythos as a technology that can move between government actors faster than regulators can settle on a common approach, underscoring how hard it is for institutions to align on what “safe enough” looks like.

The exclusion matters beyond a single vendor because it signals a practical policy outcome: even if a model is capable and potentially valuable, access can be denied if the using agency believes it cannot confidently control downstream uses in classified settings.

At the same time, the DOD’s other partnerships show the direction of travel: the Pentagon is still willing to deploy advanced AI tools across its stack, including among companies such as OpenAI, Google, Nvidia, and others, under a “lawful use” framing.

Why this is a big deal

For defense contractors and model providers, the Anthropic decision is a reminder that security isn’t only a technical barrier—it’s also a governance and deployment question. The contracts suggest that vendors may need clearer compliance pathways, stronger constraints, or demonstrated controls to access the most sensitive environments.