Trump bans Anthropic, but federal agencies are still secretly testing the Mythos model.

```

Despite the White House’s explicit ban on cooperating with Anthropic, multiple federal agencies are quietly circumventing this prohibition to test the company’s newly released high-risk AI model—Claude Mythos. This contradictory situation reflects the increasingly sharp tension in the Trump administration between blocking allied tech firms and protecting national cybersecurity.

According to media reports citing insiders, the AI Standards and Innovation Center (CAISI) under the Department of Commerce is actively testing Mythos, evaluating its cyber offense and defense capabilities as well as potential national security risks. Meanwhile, at least two major federal agencies have proactively contacted Anthropic, expressing interest in integrating Mythos into their cybersecurity defense efforts. It is reported that IT officials from the Treasury Department are also seeking to use Mythos to patch unknown vulnerabilities in the agency’s network.

Mythos’ capabilities have both impressed and worried researchers—the model can discover critical software vulnerabilities that even top human experts cannot identify. Within just a week of Mythos being publicly announced by Anthropic, staff from at least three Congressional committees have sought or already received briefings from Anthropic regarding its cyber scanning capabilities. Former national security officials warn that if this contradiction is not properly managed, it will seriously impair America’s cybersecurity posture.

Ban Origins: Amodei Refuses to Let AI Participate in “Autonomous Lethal Attacks”

The conflict between the Trump administration and Anthropic originated from a policy clash in February this year. Anthropic CEO Dario Amodei explicitly refused to allow the Pentagon to use the company’s models for autonomous lethal attacks or mass surveillance on American citizens. In response, Trump and Secretary of Defense Pete Hegseth immediately ordered all federal agencies to stop using Anthropic’s technology.

Hegseth subsequently formally listed Anthropic as a “supply chain risk,” an unprecedented move by the U.S. government against a domestic tech firm that effectively blocked Anthropic’s models from being used in Defense Department contracts. Trump himself called the company’s management “left-wing lunatics” on social media, citing their opposition to using its tools for domestic mass surveillance or autonomous lethal strikes.

Anthropic has filed lawsuits against the supply chain risk designation, and due to federal law’s special provisions, separate cases have been filed in two courts. Currently, rulings are divided: The Northern District Federal Court in California has suspended part of the governmental designation, but the D.C. Circuit Court of Appeals shortly after temporarily upheld it.

CAISI’s Secret Testing: “Red Team” Evaluations Under the Ban

Despite the ban still hanging, CAISI—an agency under the U.S. National Institute of Standards and Technology (NIST), founded in 2024 with its name changed last year by the Trump administration—reportedly began testing Mythos’ offensive capabilities even before Anthropic’s public announcement. According to media citing four sources, CAISI researchers are currently conducting “red team” evaluations of Mythos to gauge its capabilities and possible national security risks.

NIST did not respond to requests for comment. An anonymous Anthropic official confirmed that the company has provided Mythos to the government for “independent testing and technical evaluation,” but no federal agency is publicly named in Anthropic’s announcement.

Charlie Bullock, senior researcher at the think tank “Law and AI Research Institute,” pointed out the situation's irony: “The U.S. government tries to prohibit government agencies from using Anthropic’s products—and just weeks later, the company launches a revolutionary product crucial to cybersecurity and national security.” He further noted that if the California court sides with the government, federal agencies may not be permitted to test Mythos at all.

Chilling Effect: Political Pressure Hinders Deep Collaboration

Although some agencies are quietly pushing ahead with testing, White House political pressure continues to produce a substantive “chilling effect.” A former senior national security official said that the Trump administration’s public stance prevents federal agencies from openly engaging with Anthropic or overtly deploying its model to detect and fix cybersecurity vulnerabilities—such tasks require large-scale software engineering teams and substantial resources, making it easy to be labeled as violating presidential directives.

Congress also voices dissatisfaction. Three congressional aides responsible for AI policy stated that the White House’s position prevents the government from fully leveraging the nation’s most advanced technologies. One aide bluntly remarked, “The Pentagon, by flipping off the most capable AI vendor, has shot itself in the foot.”

In a statement, the White House said the Trump administration “continues to work with AI companies to ensure their models help protect critical software vulnerabilities,” and claimed the government is “actively coordinating with government and industry to ensure the United States and Americans are protected.” Pentagon spokespersons declined to comment.

Bleak Outlook: Adversaries May Achieve Equivalent Capabilities Within Two Years

Anthropic states that Mythos is currently only available to select tech and cybersecurity organizations, citing its ability to discover and exploit unknown software vulnerabilities as being too risky for public release. The company further predicts that models with equivalent cyberattack capabilities will become widespread within the next two years, potentially triggering a wave of large-scale novel cyberattacks.

Some former national security officials are deeply concerned by this prospect. Former NSA legal counsel Glen Gerstell stated:

“I do hope the current tensions between the Pentagon and Anthropic will not hinder work that’s critical to cybersecurity.”

Deputy Director of the CIA Michael Ellis continued the White House’s tough stance in a speech last Friday, saying the CIA “will not allow private companies to dictate how and when the agency legally uses their technology.” In Hegseth’s earlier directive, Defense Department officials were given six months to complete a “smooth transition” from Anthropic’s products to “better” alternatives.

Risk Warning and DisclaimerThe market carries risks and investments require caution. This article does not constitute personal investment advice, nor does it consider the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are appropriate for their particular circumstances. Investing based on this information is at your own risk. ```