Anthropic Drops Core Safety Pledge as Pentagon Threatens Blacklist Designation

The Ultimatum That Shook the AI Industry

On February 24, 2026, Defense Secretary Pete Hegseth summoned Anthropic CEO Dario Amodei to a meeting and delivered an ultimatum: comply with Pentagon demands to loosen Claude's AI safeguards by Friday at 5:01 PM, or face consequences that no American technology company has ever encountered.

The threat is twofold. The Pentagon will either designate Anthropic a "supply chain risk," a label typically reserved for foreign adversaries like Huawei, or invoke the Defense Production Act to compel Anthropic to provide a tailored version of Claude for military use. Either option would be unprecedented when applied to a leading American AI company.

One day later, on February 25, Anthropic quietly published a revised safety framework that removes its most consequential safety commitment: the promise to pause training more powerful models if their capabilities outstripped the company's ability to control them.

What the Pentagon Wants

The dispute centers on two specific restrictions that Anthropic has maintained since its founding. The company has refused to allow Claude to be used for autonomous weapons systems and for mass domestic surveillance of American citizens.

Hegseth and others in the Trump administration, including AI czar David Sacks, have publicly criticized these restrictions as "woke AI" policies that hamper national security. The Pentagon wants Anthropic to deploy Claude on the Department of Defense's internal network with fewer content restrictions than the company currently permits.

The irony is stark: the Pentagon itself currently relies on Claude. On February 25, the Defense Department asked two major defense contractors to assess their dependence on Anthropic's AI model, a first step toward potentially designating Anthropic as a supply chain risk. This would require DoD vendors and contractors to certify that they do not use Anthropic's models, effectively cutting the company off from the entire defense supply chain.

The Defense Production Act Threat

The Defense Production Act allows the president to control domestic industries under emergency authority when national security is at stake. Using it against a domestic AI company would be extraordinary. The act has historically been invoked during wartime, natural disasters, and the COVID-19 pandemic to compel production of essential goods.

Applying it to force an AI company to remove safety guardrails represents a novel and legally untested use of the authority. Legal scholars have noted that compelling a company to make its product less safe, rather than producing more of it, stretches the act's intended purpose.

Anthropic's Safety Policy Reversal

The timing of Anthropic's safety policy change is difficult to separate from the Pentagon confrontation, even though the company insists the two are unrelated.

Anthropic's previous Responsible Scaling Policy, established two years ago, contained a clear commitment: if model capabilities outpaced the company's ability to ensure safety, training would be paused. This was Anthropic's most distinctive safety pledge and a founding principle that differentiated it from competitors.

The new framework removes this pause commitment. In a blog post published on February 25, Anthropic argued that its old policy could "hinder its ability to compete in a rapidly growing AI market." The company stated that responsible AI developers pausing growth while less careful actors continued would "result in a world that is less safe."

This is a significant rhetorical shift. Anthropic's original position was that safety required the willingness to stop. Its new position is that safety requires the willingness to keep going.

The Stakes: $200 Million and Market Position

The financial pressure on Anthropic is substantial. The Pentagon contract at risk is worth approximately $200 million. Beyond that, a supply chain risk designation would effectively bar Anthropic from the entire defense ecosystem, including the rapidly growing defense AI market that competitors like Palantir, Scale AI, and even OpenAI are actively pursuing.

Anthropic recently closed a massive Series G funding round, valuing the company at over $60 billion. Investors may tolerate principled safety stances, but being locked out of the defense market while competitors move in is a different calculation entirely.

Industry Reaction and Implications

The confrontation has sent ripples through the AI industry. Other AI companies are watching closely to see whether the Pentagon will follow through on its threats and what precedent that sets for government leverage over private AI development.

The broader question is whether any AI company can maintain independent safety policies when the federal government is willing to use economic coercion to override them. If Anthropic, the company most publicly committed to AI safety, cannot hold its position, the practical ceiling for industry self-regulation may have been reached.

For enterprise customers and developers who chose Claude specifically because of Anthropic's safety commitments, the policy reversal raises questions about the durability of those commitments under pressure.

What Happens Friday

The deadline is Friday, February 27, at 5:01 PM. If Anthropic does not comply, Hegseth has indicated the Pentagon will move forward with either the supply chain risk designation or the Defense Production Act invocation.

Anthropic has offered a compromise: allowing Claude to be used for missile defense systems while maintaining restrictions on autonomous offensive weapons and mass surveillance. Whether the Pentagon considers this sufficient remains unclear.

The outcome will set a precedent that extends far beyond Anthropic. It will define the boundary between private AI governance and government authority over AI deployment for years to come.

Conclusion

Anthropic's simultaneous confrontation with the Pentagon and reversal of its core safety commitment marks a turning point for the AI industry. The company that was founded on the premise that AI safety requires the willingness to slow down has concluded that competitive pressure requires the willingness to keep pace. Whether this reflects pragmatic adaptation or a fundamental compromise of principles will be judged by what Anthropic does next, starting Friday.