One Day After 'We Trust Anthropic,' the Pentagon Says the Blacklist Still Stands

On September 2, 2026, Commerce Secretary Howard Lutnick told Axios that the Trump administration’s long war with Anthropic was over. Asked whether he trusts CEO Dario Amodei, Lutnick was unequivocal: “We trust Anthropic. They’ve done what we asked. They’re back on the right side.”

The thaw lasted exactly one day.

On September 3, Emil Michael — the under secretary of defense for research and engineering, and the Pentagon’s top AI official — posted on X that Anthropic is still considered a “Supply Chain Risk” by the Department of Defense and across the entire defense industrial base. Reuters reported the remarks, and noted that the Defense Department did not respond to a request for comment. So the same executive branch, within 24 hours, told the public both that Anthropic is trusted and that it remains blacklisted.

A seven-month feud in brief

The contradiction is the latest turn in one of the strangest confrontations between the U.S. government and an American technology company in decades.

• July 2025: Anthropic and the Pentagon signed a contract making Claude the first frontier AI model approved for sensitive defense use.
• January–February 2026: The relationship ruptured over Anthropic’s safety “red lines” — usage limits the company refused to lift for military applications. Reporting later traced tension to Claude’s role in operations connected to the Iran conflict. Defense Secretary Pete Hegseth publicly rebuked Amodei, who said the company “cannot in good conscience accede” to the Pentagon’s demands.
• February 27, 2026: Hegseth moved to designate Anthropic a “supply chain risk,” and President Trump directed federal agencies to cease using Anthropic technology. The same day, the Pentagon struck a deal with OpenAI — a competitor substitution that raised eyebrows then and looks pointed now.
• March 5: The designation became formal. Anthropic became the only American company in history to be publicly named a supply chain risk — a label traditionally reserved for foreign adversary suppliers.
• March 9: Anthropic sued in two federal courts.
• March 26: A federal judge in San Francisco granted a preliminary injunction, citing what CNBC described as “First Amendment retaliation” — the government punishing a company for criticism.
• April 8: An appeals court denied Anthropic’s emergency stay request relating to the separate FASCSA (Federal Acquisition Supply Chain Security Act) track, leaving that designation in effect for covered systems while litigation continued.
• August 27: A federal judge ruled definitively for Anthropic, finding the Pentagon had illegally punished the company for criticizing the government. The judge called the decision “illegal and baseless” and struck down the designation.
• September 2–3: Lutnick’s olive branch, then Michael’s X post pulling the rug out.

What “still in effect” actually means

The Pentagon’s position is not legally frivolous — and that is what makes this moment significant. The August 27 ruling struck down the department’s core designation, but the FASCSA track and the pending appeal give the government procedural room to argue that, until appeals are exhausted, the flag stays up in procurement systems.

The practical effect falls hardest on contractors. Even a court order saying a label is “illegal and baseless” doesn’t automatically scrub it from every procurement database, clause, and compliance checklist across the defense industrial base. Contractors are conservative by instinct and by contract law; while any formal “risk” flag exists, buying Anthropic capacity carries perceived legal exposure. That is precisely the chilling effect a blacklist is designed to produce — and it persists even in defeat. Palantir, for one, spent months unwinding Claude from the Pentagon’s AI software stack after the original designation.

The split brain problem

The deeper story is that nobody in this administration appears to be in charge of AI policy toward a single company. The Commerce Department — which controls export controls and has every incentive to project that American AI firms are stable, trustworthy partners — says the feud is over. The Defense Department, where the designation originated and where officials staked personal credibility on it, says nothing has changed.

Emil Michael’s role deserves emphasis. He is the same Pentagon AI chief who reported his third round of stock sales this summer, drawing ethics scrutiny, and he is now the official single-handedly sustaining a blacklist that a federal judge has already struck down once. Whether one reads his X post as bureaucratic self-protection, inter-agency score-settling, or genuine institutional conviction about supply chain risk, the result is the same: the executive branch is publicly contradicting itself about whether a major American AI company is trusted.

For Anthropic, the material damage is contained but real. The company’s commercial business is thriving — Claude Fable 5.1 shipped September 1 across AWS, Google Cloud, and Azure — and its enterprise momentum doesn’t depend on defense contracts. But the U.S. defense market is one of the largest AI buyers on earth, and while the flag flies, rivals fill the void. Reuters reported in April that the Pentagon’s ouster of Anthropic was already “opening doors” for smaller AI firms willing to operate without red lines. OpenAI, which took the Pentagon deal Anthropic lost, has since positioned itself as the patriotic default — its billion-dollar cyberdefense commitment announced this week lands very differently in that context.

Why this matters beyond Anthropic

Strip away the personalities and two precedents remain.

First, the First Amendment retaliation finding. A court has now formally held that the government may not use procurement tools to punish a company for speech — in this case, Anthropic’s public criticism of how its models were being used. If that holding survives appeal, it is one of the most important AI-governance rulings to date: it draws a line the state may not cross even when it invokes national security.

Second, the enforcement question. If a department can keep a judicially invalidated designation in operation across the industrial base — through appeals, parallel statutory tracks, and public statements — then the ruling’s force is largely symbolic for as long as the appeals take. That is a live test of whether court orders bind the executive in practice, not just in theory, in the AI era.

Every frontier lab is watching. Anthropic’s original sin, in the Pentagon’s telling, was insisting on red lines its competitors did not. The lesson the market is drawing is uncomfortable: safety constraints on government use may be legally protected but operationally punished, while compliance gets rewarded with contracts. How the White House resolves its own contradiction — by forcing DoD to formally rescind, or by letting the flag fly through the appeals — will tell the entire industry which incentive actually governs.

For now, the official answer to “does the United States trust Anthropic?” is: yes, no, and it depends which building you ask.

Sources

• Reuters (via SRN News): “Anthropic still flagged as risk to defense industrial base, US official says” — Mike Stone, Sept 3, 2026
• Reuters: “US Commerce chief says Anthropic back on ‘right side’ with Trump administration” — Sept 2, 2026
• Axios: “Lutnick: Anthropic is ‘back on the right side’ with Trump administration” — Sept 2, 2026
• AP News: “Judge rules in favor of Anthropic in case against Pentagon” — Aug 28, 2026
• CNBC: “Anthropic wins preliminary injunction in Trump DOD fight as judge cites ‘First Amendment retaliation’” — Mar 26, 2026
• Forbes: “Anthropic Back On Trump’s ‘Right Side,’ Lutnick Says” — Sept 2, 2026
• Wikipedia: “Anthropic–United States Department of Defense dispute” — timeline and litigation record