# Anthropic v. Department of War (D.C. Cir., Sept. 25): the declarations Anthropic filed to prove it could not sabotage… *Canonical page:* https://truthscore.com/source/a4219cc2 *Machine twins:* https://truthscore.com/source/a4219cc2.md · JSON at https://truthscore.com/api/v1 Publisher: reddit.com · published 2026-09-26 · read 2026-09-27 Original: https://www.reddit.com/r/law/comments/1wqijta/ *A source is testimony, not proof. The document itself is never republished here.* ## Who is named (6) - [FASCSA](https://truthscore.com/entity/0951fd57) - [Claude](https://truthscore.com/entity/2c649661) - [U.S. Department of War](https://truthscore.com/entity/6d0485f2) - [Anthropic](https://truthscore.com/entity/83262514) - [Judge Lin](https://truthscore.com/entity/a4b5864b) - [D.C. Circuit](https://truthscore.com/entity/a87a2b10) ## What it added - [The panel used Anthropic's declarations as evidence that Anthropic would manipulate Claude.](https://truthscore.com/claim/e4ea3ba9) > Anthropic v. Department of War (D.C. Cir., Sept. 25): the declarations Anthropic filed to prove it could not sabotage Claude are the evidence the panel used to hold it would "manipulate" it. - [Anthropic has no access to air‑gapped deployments.](https://truthscore.com/claim/1ebd2e52) > Anthropic's head of public sector attested that it has no access to air‑gapped deployments - [Delivered models do not drift.](https://truthscore.com/claim/c87eb191) > that delivered models do not drift - [Model training is the primary mechanism through which Anthropic can influence the behavior of models used by the Department.](https://truthscore.com/claim/b210d5d3) > "model training is the primary mechanism through which Anthropic can influence the behavior of models used by the Department." - [In San Francisco the evidence was unrebutted and dispositive, showing there was no back door, no kill switch, no sabotage, and no adversary.](https://truthscore.com/claim/61468e14) > In San Francisco that evidence was unrebutted and dispositive: no back door, no kill switch, no sabotage, no adversary. - [In Washington the same sentences established that Anthropic deliberately shapes the design and operation of the product so that it refuses two categories of…](https://truthscore.com/claim/66b4a516) > In Washington the same sentences established that Anthropic deliberately shapes the design and operation of the product so that it refuses two categories of tasks. - [Under § 4713 manipulation is a certainty that turns on what Anthropic does, not why Anthropic does it.](https://truthscore.com/claim/af92223c) > Under § 4713 that is "not only a 'risk' but a certainty" of manipulation, and the definition "turns on what Anthropic does, not why Anthropic does it." - [The Department's memo said an Anthropic executive questioned the propriety of a contractor's use of Claude for a sensitive military operation abroad.](https://truthscore.com/claim/8fd52352) > The Department's memo said only that an Anthropic executive "questioned the propriety" of a contractor's use of Claude "for a sensitive military operation abroad." - [Anthropic added press reports to the record identifying the operation as the January 3 operation in Venezuela.](https://truthscore.com/claim/b1392b53) > Anthropic, presumably to show how trivial the episode was, put press reports in the record identifying it as the January 3 operation in Venezuela. - [The panel responded that the possibility of a kinetic operation to capture a foreign head of state underscores the fraught nature of the dispute and the…](https://truthscore.com/claim/14d13ef0) > The panel's response: the possibility that this happened during a "kinetic operation to capture a foreign head of state" only "underscores the fraught nature" of the dispute and the Department's need to know its systems will work. - [The Department declared that models with five to ten trillion parameters make rigorous analysis or auditing of outputs mathematically impossible.](https://truthscore.com/claim/16b5859f) > The Department declared that models with five to ten trillion parameters make "rigorous analysis or auditing" of outputs "mathematically impossible." - [Anthropic conceded some legitimacy to the Department's concern about the opacity of these systems generally.](https://truthscore.com/claim/5ad0e64f) > Anthropic, rather than fight the science, conceded "some legitimacy to DoW's concern about the opacity of these systems generally." - [The court closed the testing argument, stating testing is no panacea, refusals vary with wording, and the Department cannot utilize AI systems that remain…](https://truthscore.com/claim/0f7915c2) > The court took the concession and closed the door on the testing argument: testing is no panacea, refusals vary with wording, and the Department "cannot utilize AI systems that remain trapped in amber." - [FASCSA requires notice and an opportunity to respond before exclusion, absent urgency.](https://truthscore.com/claim/1157f0a6) > FASCSA requires notice and an opportunity to respond before exclusion, absent urgency. - [Anthropic filed a full rescission request on April 17 and lost on June 3.](https://truthscore.com/claim/82ae7377) > Anthropic got neither, then filed a full rescission request on April 17 and lost on June 3. - [The court cited reported investment offers above $900 billion and questioned whether Anthropic has suffered any such harm.](https://truthscore.com/claim/3552d521) > The court cited reported investment offers above $900 billion and said "one may fairly question whether Anthropic has suffered any such harm," adding that Anthropic never explained why a termination for the same stated reasons would carry less stigma. - [Anthropic won the statute it wrote the record for.](https://truthscore.com/claim/4240495b) > Anthropic won the statute it wrote the record for. - [Lin issued an August order.](https://truthscore.com/claim/05c1f971) > Opinion, Lin's August order and the directive in the first comment. ## What it cost Reading this document into the graph cost $0.02.