Pentagon CTO Says Anthropic Remains Barred as Mythos Raises Security Stakes

Emil Michael, the Pentagon’s chief technology officer, refuted recent reports suggesting a softened relationship between the Department of Defense and Anthropic. While the capabilities of Anthropic’s Claude Mythos have attracted government interest, no renewed cooperation has been established.

"Mythos’s vulnerability is a national‑security issue. We must ensure our cybersecurity because the model can discover and fix vulnerabilities," Michael said.

Claude Mythos is one of Anthropic’s latest models, part of the broader Claude system that includes AI assistants comparable to OpenAI’s ChatGPT and Google’s Gemini. Anthropic released it in early April as a “Mythos Preview.”

Red‑team researchers who tested the model reported that Mythos exhibits “astonishing ability” on computer‑security tasks, quickly locating hidden vulnerabilities in decades‑old code and demonstrating how easily those flaws can be exploited.

Rather than making Claude widely available, Anthropic limited access to the Mythos tool to twelve companies through a program called “Glass Wing,” describing the effort as a way to protect the world’s most critical software.

CNBC’s Becky Quick noted that Anthropic is considered a supply‑chain risk and that the DoD does not wish to deploy Anthropic systems. Some federal agencies have accessed Mythos, but only for evaluation purposes, not for production deployment.

Media reports indicated that the National Security Agency began using Mythos, fueling rumors that the government’s stance had softened. Earlier, the Pentagon had disputed Anthropic’s standard‑use policies, which further fueled speculation about broader adoption. However, Michael reaffirmed that the official position remains unchanged: federal agencies are merely analyzing Mythos.

Michael added that the government continues to explore how to work with companies that develop frontier models, including emerging tools such as ChatGPT 5.5‑Cyber. He cited upcoming meetings between AI leaders and the White House to discuss the cybersecurity risks posed by Mythos and similar models.

Looking ahead, Michael said, “We think Anthropic may be the first company eliminated, but other companies will certainly appear.” He emphasized that the government intends to collaborate with these firms over the next year to understand their capabilities and address any problems that arise in both the private and public sectors.