The Pentagon Blacklisted Anthropic While Claude Was Selecting Targets in Iran

On February 28, 2026, the U.S. military used Anthropic’s Claude AI to help select and prioritize over 1,250 targets during Operation Epic Fury, the coordinated U.S.-Israeli strike campaign against Iran. One day earlier, on February 27, the Pentagon had blacklisted Anthropic — designating it a “supply chain risk to national security,” a penalty previously reserved for adversarial foreign entities like Huawei.

The $200 million contract collapsed. Defense Secretary Pete Hegseth ordered every defense contractor, supplier, and partner to sever all commercial ties with the company whose AI had just helped run the largest military operation in the Middle East since the 2003 invasion of Iraq. The contradiction isn’t subtle — and the dispute that triggered the blacklist wasn’t about capability, reliability, or national security. It was about five words in a contract and $26 million in political donations that Anthropic never made.

Claude Was Selecting Targets While the Pentagon Blacklisted Anthropic

Two sources confirmed to CBS News that Claude was active inside Palantir’s Maven Smart System during the February 28 Iran strikes — and continued operating afterward. Claude’s Pentagon deployment began in late 2024, when Anthropic partnered with Palantir to integrate its model into the military’s primary intelligence analysis and targeting platform.

Claude’s tasks during Operation Epic Fury weren’t administrative. The AI processed classified satellite, surveillance, and signals intelligence data. It ran war-game simulations. It identified targets, generated precise location coordinates, and prioritized those targets according to importance. The U.S. struck over 1,250 Iranian targets in the first 48 hours — a volume of targeting decisions that exceeded human analytical capacity at that pace.

This is the most advanced confirmed AI deployment in active warfare to date. And it happened under a vendor the government was simultaneously calling a national security threat. CBS News reported the Pentagon could need three months or more to replace Claude’s capabilities — a strange timeline for something classified as an urgent risk.

The Designation That Makes No Legal Sense

Hegseth invoked 10 U.S.C. Section 3252 — a statute designed for adversarial foreign entities, never previously applied to an American company. The designation didn’t just cancel Anthropic’s contract. It barred every defense contractor, supplier, and partner from any commercial activity with the company. President Trump reinforced the message on Truth Social, calling Anthropic a “RADICAL LEFT, WOKE COMPANY.”

The government’s position contains a contradiction that legal scholars noticed immediately. Weeks before the blacklist, the Pentagon had considered invoking the Defense Production Act to compel Anthropic to provide access to Claude — arguing it was indispensable to national defense. Then it turned around and designated that same technology a grave supply chain risk requiring government-wide removal.

Lawfare scholars Michael Endrias and Alan Rozenshtein were direct: “The government is arguing that Claude is so vital to military operations that it cannot tolerate any contractual restrictions on it — while simultaneously claiming that Claude poses such a grave supply chain risk that the entire federal government must stop using it.” Independent analyses from Just Security and Mayer Brown reached the same conclusion. Lawfare’s verdict: “The designation won’t survive first contact with the legal system.”

Five Words That Revealed the Pentagon’s Real Intent

Most coverage frames the dispute as a broad AI ethics standoff. It wasn’t. Anthropic CEO Dario Amodei’s 1,600-word internal staff memo — leaked and reported by The Information — reveals the contract collapsed over a single phrase.

Near the end of negotiations, the Pentagon offered to accept all of Anthropic’s terms if the company deleted one specific clause: “analysis of bulk acquired data.” Amodei wrote that this “exactly matched this scenario we were most worried about” — mass surveillance using AI to assemble scattered personal data into profiles at scale. When someone agrees to everything except the one restriction that would prevent a specific abuse, that tells you what they were planning.

Anthropic’s official statement made the concern explicit: “Powerful AI makes it possible to assemble scattered data into a comprehensive picture of any person’s life — automatically and at massive scale.” Amodei rejected the Pentagon’s ultimatum, stating: “We cannot in good conscience accede to their request.” Five words in a contract clause. A $200 million deal. The closest thing to a confession available in any public record.

OpenAI Got the Same Terms Anthropic Was Punished For Demanding

Hours after the blacklist, Sam Altman announced OpenAI’s classified Pentagon deal — which includes prohibitions on mass domestic surveillance and fully autonomous weapons. Both restrictions Anthropic had demanded. Both restrictions Anthropic had been punished for insisting upon.

Amodei’s memo didn’t mince words. He called OpenAI’s deal “straight up lies” and “safety theater,” writing: “The main reason they accepted and we did not is that they cared about placating employees, and we actually cared about preventing abuses.”

The political math isn’t subtle. OpenAI President Greg Brockman and his wife donated $25 million to the MAGA Inc super PAC. Sam Altman donated $1 million to Trump’s inauguration. Anthropic donated nothing. Amodei identified this in his memo, alleging that “the real reasons DoW and the Trump admin do not like us is that we haven’t donated to Trump.”

Meanwhile, Elon Musk’s xAI signed a $200 million deal and agreed to the “all lawful purposes” standard without negotiation. The compliant path is now the safe path, and every AI company is watching.

The Chilling Effect Is Already Working

Lockheed Martin pledged compliance and began removing Claude. Benzinga reported that Palantir faces a costly overhaul of Maven Smart System, where Claude was integrated into core workflows — Maven-related contracts potentially exceed $1 billion. One defense tech investor told reporters that 10 portfolio companies “backed off of their use of Claude for defense use cases and are in active processes to replace the service.” Not because Claude failed. Because association with Anthropic is now a business risk.

The backlash cut both directions. ChatGPT uninstalls surged 295% after OpenAI’s deal announcement, while Claude hit #1 on the U.S. App Store. Nearly 900 Google and OpenAI employees signed the “We Will Not Be Divided” open letter, recognizing the government’s divide-and-conquer strategy: “They’re trying to divide each company with fear that the other will give in. That strategy only works if none of us know where the others stand.”

But the chilling effect doesn’t need the designation to survive legal challenge. It just needs other companies to decide the risk isn’t worth it. When 10 defense-adjacent startups drop a vendor not because of performance but because of political association, the designation has already accomplished its goal — regardless of what any court eventually rules.

Loyalty Tests, Not Policy Disputes

If Anthropic wins in court and the designation is struck down — which Lawfare’s analysis suggests is likely — does that actually restore the industry norm of negotiating ethical limits? Or does it just teach the next administration to use a cleaner pretext?

The Pentagon didn’t blacklist Anthropic for refusing to enable mass surveillance. It blacklisted Anthropic for refusing to pretend it wouldn’t. OpenAI’s deal includes the same restrictions on paper — but it came with $26 million in political donations and a willingness to let the Pentagon define what “safeguards” mean in practice. That’s the new standard. Not what your contract says, but whether you paid for the privilege of having one.

As of March 5, Amodei and Pentagon Undersecretary Emil Michael are back at the negotiating table. The White House has separately cast doubt on reconciliation. Whether Anthropic files its promised legal challenge — and whether the court precedent that follows reshapes or destroys AI ethics clauses in federal contracts — will be determined in the weeks ahead.