Pentagon Upholds Anthropic Blacklist as White House Seeks Access to Mythos AI

The Trump administration's self-proclaimed hands-off approach to artificial intelligence is colliding with the realities of advanced technology national security demands and entrenched bureaucratic resistance exposing deep contradictions in its America First agenda just months into the president's second term.

At the center of the turmoil sits Anthropic a company the Pentagon formally labeled a supply chain risk earlier this year a designation typically reserved for Chinese or Russian entities. That unprecedented step followed the collapse of negotiations over how the Defense Department could safely use Anthropic's models in classified environments. Lawsuits public recriminations and alternative contracts with rival AI firms followed. Yet the same administration now finds itself quietly thawing relations with the company it once sought to sideline.

The reason is as urgent as it is uncomfortable. Anthropic's latest model Mythos possesses sophisticated capabilities for identifying cyber vulnerabilities and suggesting patches. Government agencies despite the Pentagon's ongoing legal battles have begun testing Mythos alongside other frontier systems. Emil Michael the Defense Department's chief technology officer nominee acknowledged the bind in an interview Friday describing Mythos as a separate national security moment requiring hardened networks while confirming that Anthropic itself remains blacklisted.

This awkward dance reflects a broader policy collapse. Candidate Trump promised to slash regulations unleash American innovation and prevent China from dominating the AI race. In practice the administration is discovering that the most powerful models cannot simply be wished away or easily replaced. Axios reporting reveals internal discussions about an executive order that would have purged Anthropic systems entirely from government use. That plan has been shelved as the White House recognizes the technology's value even as it continues fighting the company in court.

The situation is further complicated by regulations rushed through in President Biden's final days. In the last week of his term the Bureau of Industry and Security within the Commerce Department implemented the so-called AI Diffusion Rule. The measure required domestic AI chip manufacturers to reserve half their production for U.S. consumption severely restricting exports even to close allies. Trump promptly rescinded the rule calling it ill-conceived. Yet according to critics inside and outside the administration Biden-era holdovers and career officials have slowed efforts to fully unwind it.

Conservative commentators describe this as classic deep state sabotage. They argue the restrictions were designed to undermine the incoming administration and that they remain ineffective in any case. A Japanese buyer for instance could legally purchase restricted chips and transship them to China with little to stop the flow. Defenders of the original rule claimed they were simply closing loopholes before the transition. The result is a policy vacuum where rhetoric about unleashing American AI meets the grinding machinery of inherited regulations and risk-averse bureaucracy.

This is not abstract regulatory squabbling. The stakes involve who controls the most powerful computational tools on the planet and how those tools are deployed in both commercial and military contexts. The Trump administration entered office vowing minimal government interference in the AI sector. Yet as models like Mythos demonstrate capabilities that blur the line between defensive cyber tools and offensive weapons the White House is being dragged into precisely the kind of intervention it campaigned against.

By regulating through contracts and procurement decisions individual agencies are accumulating enormous unchecked power. When one company becomes both indispensable and politically toxic the resulting policy whiplash serves no one's interest least of all the goal of maintaining American technological edge over strategic competitors.

The Anthropic episode also raises uncomfortable questions about the national security state's ability to separate genuine risk from bureaucratic score-settling. Labeling an American AI firm a supply chain threat sets a dangerous precedent that could chill innovation or be weaponized against future disfavored companies. At the same time ignoring the genuine security implications of models that can probe network weaknesses at machine speed would be reckless.

What emerges from reporting across outlets is a portrait of an administration caught between its ideological commitments and technological reality. Trump officials want to project decisiveness and deregulation. The facts on the ground show confusion court battles legacy rules that refuse to die and a growing realization that some AI systems are becoming too powerful to treat as ordinary commercial products.

As agencies continue testing Mythos and similar models behind the scenes the White House faces a choice. It can persist with public posturing about bureaucratic sabotage while quietly relying on the very companies it has attacked. Or it can develop a coherent strategy that acknowledges both the transformative potential of these systems and the legitimate need to prevent them from being misused or leaked to adversaries.

So far the evidence points to the former. The result is not the bold American AI renaissance that was promised but a messy rearguard action managed by the same permanent government the president pledged to dismantle. In the race to dominate artificial intelligence such internal contradictions are not merely political embarrassments. They are strategic liabilities that neither deep state narratives nor corporate press releases can fully obscure.