Anthropic's Mythos Captivates Washington Despite Pentagon Blockade

Anthropic and the White House: A Strategic Meeting

During our coverage of the Glasswing project earlier this month, the story focused on a model deemed too dangerous to be made public and what Anthropic decided to do about it. This story has evolved. On Friday, Anthropic CEO Dario Amodei entered the West Wing for a meeting with White House Chief of Staff Susie Wiles. Treasury Secretary Scott Bessent was also present. The White House described the discussions as "productive and constructive," which Anthropic confirmed. When a reporter asked President Trump what he thought of the visit on a tarmac in Phoenix, he responded, "Who?" and stated that he had "no idea" Amodei was there.

Aside from this detail, the meeting itself represents one of the most striking political turnarounds in the recent history of AI. Just a few weeks ago, the Trump administration had declared that Anthropic posed a risk to the supply chain—a designation typically reserved for foreign adversaries—and Trump himself claimed that the administration "would no longer do business with them." A federal judge in San Francisco has since blocked the enforcement of that directive, allowing Anthropic to work with non-military agencies while the litigation unfolds. The dispute with the Pentagon remains very active.

The Model and the Politics

What has changed the game—at least at the White House level—is Anthropic's Mythos AI cybersecurity capability. More specifically, the fact that agencies are reportedly monitoring Mythos to achieve things that no other tool can do, and they are not willing to sit idly by. As we reported when Anthropic unveiled the Glasswing project, Mythos Preview was not specifically trained for security work. Its ability to autonomously identify and exploit software vulnerabilities emerged from general improvements in reasoning and coding, and what it has discovered since its deployment has been striking.

During internal testing, Mythos located thousands of unknown and high-severity vulnerabilities across all major operating systems and web browsers, including a 27-year-old bug in OpenBSD and a 16-year-old flaw in FFmpeg that had passed automated tests five million times without detection. Rather than making it public, Anthropic has only shared it with a select group of organizations through the Glasswing project—a coalition that includes AWS, Apple, Cisco, Google, Microsoft, Nvidia, CrowdStrike, and JPMorgan Chase, among others—backed by up to $100 million in usage credits. The model is used offensively, in a controlled manner: to find vulnerabilities before someone else does.

The U.S. government has observed this coalition in action and wishes to participate. Intelligence agencies and the Cybersecurity and Infrastructure Security Agency are already testing Mythos, and the Treasury Department has also expressed interest, according to Axios. The Treasury and other government agencies have shown interest in joining the Glasswing list, and before Friday's White House meeting, two sources told Axios that an agreement in this direction could be reached soon.

The State of the Status Quo

In a separate Axios report, a concern raised is that Mythos and other cutting-edge AI tools could enable hackers to penetrate the U.S. financial system. Conversely, the report estimated that businesses and government agencies could use Mythos to bolster their cybersecurity defenses before malicious actors gain access to their systems. This dual-use tension is now a political issue. National Cyber Director Sean Cairncross is set to lead a group of federal agents to identify security vulnerabilities in critical infrastructures and strengthen government systems against exploitation by AI.

Friday's meeting was orchestrated to separate two conversations that had become intertwined. Before the session, both parties sought to isolate the conflict with the Pentagon from how the rest of the government interacts with Anthropic, and the next steps are expected to involve other departments' access to Mythos Preview, according to sources close to the negotiations.

A Trump advisor told Axios: "This is a big deal. Everyone is complaining. There’s all this drama. So, it was elevated to Susie to listen to Dario, determine what’s fluff, and start charting a path forward." An administration official succinctly summarized the current dynamic: "There is progress with the White House. There is no progress with the [Department of] Defense." This division is telling. Civilian agencies like the Departments of Energy and Treasury are responsible for protecting critical sectors, such as the power grid and the financial system.

Their concerns do not revolve around autonomous weapons or surveillance. They want the capability that Mythos offers, and they are not willing to be collateral damage in a conflict between the Pentagon and an AI company. The DOD has not commented on Mythos but continues to use Anthropic's Claude models in the war against Iran. This footnote deserves to be highlighted.

Publicly, Anthropic has also made moves signaling that it understands how Washington operates. Public filings show that Anthropic has recently engaged the lobbying firm Ballard Partners—where Wiles worked for years—specifically to advocate for acquisitions by the Department of Defense.

The litigation is not over. A federal appeals court has rejected Anthropic's request to temporarily block the Pentagon's blacklist; a judge in San Francisco granted a preliminary injunction in a separate case. Anthropic remains barred from contracts with the DoD but can continue to work with the rest of the government while both cases proceed.

The White House has stated that it plans to continue the dialogue with Anthropic and other AI companies, and the Office of Management and Budget is already preparing to give agencies access to Mythos to assess their defenses, according to Bloomberg. This is significant progress, even as the Pentagon remains the unresolved piece.

A source close to the negotiations expressed this clearly: "It would be grossly irresponsible for the U.S. government to deprive itself of the technological advancements that the new model presents. It would be a gift to China." This framework—less focused on Anthropic's legal situation, more on what the U.S. cannot afford to lose—is what brought Amodei to the West Wing. Whether the Pentagon will eventually follow is another question.