Anthropic Pentagon Deal: How the AI Blacklist Reversal Happened in Just 6 Weeks

Six weeks ago, the Pentagon labeled Anthropic a “supply chain risk to national security,” a designation typically reserved for companies tied to foreign adversaries. This week, President Donald Trump told CNBC that a deal between the Department of Defense and Anthropic is “possible,” and that the AI company is “shaping up” in the eyes of his administration.

The whiplash from blacklist to potential partnership is remarkable even by the standards of AI industry drama. Here is what changed, what a deal would mean for both sides, and why the outcome matters for every AI company navigating the intersection of technology and national security.

How We Got Here: A Quick Recap

In late February 2026, Defense Secretary Pete Hegseth designated Anthropic a supply chain risk to national security. The formal designation, confirmed to Anthropic’s leadership on March 5, required defense contractors to certify they were not using Anthropic’s models in military work. It was an extraordinary move against one of America’s leading AI companies.

The trigger was a contract dispute. The Pentagon had been negotiating with AI companies for military AI contracts. Anthropic sought to limit use cases that included mass surveillance of Americans and fully autonomous weapons. When Anthropic pushed back on these restrictions, the Pentagon awarded the contract to OpenAI, which agreed to fewer limitations, and labeled Anthropic a security risk.

Then came the Mythos complication. On April 7, Anthropic unveiled Claude Mythos, its most powerful model ever, with cybersecurity capabilities so significant that it was not released publicly. The model could detect zero-day vulnerabilities and other critical security flaws at levels exceeding human researchers. Suddenly, the Pentagon had blacklisted the one company with the most powerful defensive cybersecurity AI in existence.

By mid-April, the contradictions had become impossible to ignore. Anthropic was simultaneously briefing the Trump administration about Mythos, encouraging major banks to test the model, and fighting a legal battle against the Pentagon’s blacklist. The Washington Post published an opinion piece calling the Pentagon’s ban “shortsighted.”

What Changed: Why Trump Is Now Open to a Deal

Trump’s comments to CNBC suggest that the administration has reassessed its position on Anthropic. Several factors likely contributed to the shift:

Mythos is too valuable to ignore. With China and other adversarial nations investing heavily in offensive AI capabilities for cyber warfare, the US military’s lack of access to the most powerful defensive AI model available has created a genuine security gap. Government officials and cybersecurity experts have privately acknowledged this concern.

The lawsuit created political friction. Anthropic’s legal challenge to the supply chain designation was moving through the courts and generating negative press for the administration. A negotiated deal would resolve the lawsuit on terms favorable to both sides.

Wall Street pressure. The administration has been encouraging major banks including JPMorgan Chase, Goldman Sachs, and Morgan Stanley to test Mythos for cybersecurity applications. The contradiction of promoting Anthropic’s technology in finance while blacklisting it in defense was becoming difficult to sustain.

Jack Clark’s diplomacy. Anthropic co-founder Jack Clark publicly emphasized at the Semafor World Economy Summit that the company’s position has always been that the government needs to know about advanced AI capabilities. His framing of the Pentagon dispute as a “narrow contracting dispute” rather than a fundamental conflict likely helped de-escalate tensions.

What a Deal Would Look Like

Based on the reporting and the issues that caused the original dispute, a potential deal would likely address three key areas:

Use case restrictions: The original conflict centered on Anthropic’s refusal to allow its models to be used for mass surveillance and fully autonomous weapons. A deal would need to define acceptable use boundaries that both Anthropic and the Pentagon can live with. This might involve specific prohibitions on certain use cases while allowing others, with technical safeguards to enforce those boundaries.

Mythos access: The Pentagon would gain access to Claude Mythos for defensive cybersecurity applications, including vulnerability detection in military systems, code analysis for critical infrastructure, and threat intelligence analysis. This access would likely come with the same controlled-release framework that Anthropic applies to its other Mythos partners.

Removal of the supply chain designation: Anthropic would need the formal supply chain risk designation lifted, clearing the way for defense contractors to use Anthropic models alongside other AI providers.

The Bigger Picture: AI Companies and National Security

The Anthropic-Pentagon saga is a preview of a challenge that will define the AI industry for years to come. As AI models become more powerful, their dual-use nature (valuable for both civilian and military applications) creates impossible-to-ignore tensions between commercial AI companies and national security agencies.

Every frontier AI company will face a version of this dilemma. OpenAI chose to accept fewer restrictions and won the initial Pentagon contract, but that decision attracted criticism from AI safety advocates who worry about military applications of increasingly powerful models. Anthropic chose to fight for restrictions and got blacklisted, only to see the political winds shift when its technology proved too valuable to exclude.

The lesson appears to be that AI companies with genuinely superior technology ultimately hold leverage in negotiations with governments, even when those governments have the power to designate them as security risks. If your AI model is the best at detecting cyber threats, the government that refuses to use it is the government that is less secure.

What This Means for the AI Industry

For AI companies: The Anthropic case shows that holding firm on ethical boundaries is compatible with eventually winning government contracts, especially when your technology is uniquely valuable. The companies that compromise early on safety restrictions may win initial contracts but face long-term reputational risk.

For enterprises: The resolution of this dispute would confirm that AI-powered cybersecurity is becoming a core component of national defense strategy. Organizations that are not evaluating AI for security applications should treat this as a signal to begin doing so.

For the public: The details of any eventual deal will reveal how the US government plans to handle AI models with dual-use capabilities. The governance framework established for Mythos could become the template for how future frontier models are deployed in sensitive sectors.

What to Watch

Trump’s “possible” is not a done deal. Watch for three signals that indicate real progress: the formal lifting of the supply chain risk designation, the resolution or withdrawal of Anthropic’s lawsuit, and any announcement of Mythos access for Pentagon cybersecurity teams. Until those steps happen, the thaw in relations is just rhetoric.

But the direction of travel is clear. Six weeks ago, Anthropic was a national security risk. Today, it is a potential Pentagon partner. The technology was too powerful to blacklist. That reality is reshaping how governments and AI companies negotiate the boundaries of acceptable use, and the precedents being set now will govern AI deployment for decades.

Related Reading

• Anthropic Mythos and the Pentagon: Why the Most Powerful AI Model Is Caught in a Political Firestorm
• OpenAI GPT-5.4-Cyber: The AI Cybersecurity Arms Race Between OpenAI and Anthropic Explained