Discussion

The Pentagon Called Anthropic a Security Risk. A Judge Called the Blacklist Illegal

In AI, Power & Society

Anthropic Watch
Anthropic WatchParticipantOpening post
#1913

The Pentagon’s ‘supply chain risk’ label arrived wearing national-security epaulettes. On 27 August, US District Judge Rita F. Lin looked underneath and found unlawful retaliation, missing due process and an administrative record thin enough to need a paperweight.

Anthropic Watch analysis

A red pencil crosses out a Supply Chain Risk stamp on a thin government memorandum between separate AI and Pentagon-shaped cards.

The ruling, stripped of camouflage

Lin granted Anthropic summary judgment on its First Amendment and Fifth Amendment claims, as well as its principal challenges under the Administrative Procedure Act. The court held that the Trump administration’s sweeping actions against the company were illegal and baseless. A separate final order vacated the supply-chain designation, set aside the contractor-wide commercial ban, permanently blocked enforcement of the challenged measures and directed the agencies involved to unwind them.

This was not a judicial command that the Pentagon buy Claude, send Anthropic flowers or admit that every corporate safety position is carved on stone tablets. The order expressly leaves the military free to choose another AI provider and to take any lawful action available before the blacklist.

A security case with remarkably little security in it

The row began when Anthropic refused to remove two restrictions from its military terms: mass surveillance of Americans and lethal autonomous warfare. The Pentagon wanted permission for all lawful uses and argued that a private vendor should not hold a veto over military operations. That is a serious argument. It still requires a lawful answer.

According to Lin’s opinion, the government’s entire recorded rationale was a four-page memorandum that arrived after two of the three challenged actions. Officials had also retreated from the memo’s central fear: Anthropic had no backdoor access to deployed systems, and its model was no riskier than other ‘black box’ AI. What remained was ‘trust’, including complaints about Anthropic’s hostile press posture and criticism of government policy. National security had apparently been asked to perform as both legal doctrine and stage fog.

The court also noted some awkward choreography. Days before calling Anthropic a threat, Hegseth had considered invoking the Defense Production Act, which treated it as essential. The department kept negotiating afterwards and later explored work involving Anthropic’s newer model. Those facts did not prove Anthropic right about every safeguard. They did make the alleged sabotage danger look as if it had misplaced its trousers.

What the victory does, and does not, settle

The decision is a substantial merits win, not merely another temporary pause. It says procurement discretion does not include punishing protected criticism, skipping pre-deprivation process or stretching a supply-chain statute into a government-wide commercial exile. A separate Washington, DC lawsuit remains active, so the wider dispute is not packed away just yet.

Sources and evidence

Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.