>10 U.S.C. § 3252 authorizes the Secretary of Defense to exclude a source from defense procurements involving national security systems if there is a supply chain risk, defined as the risk that an adversary may sabotage, maliciously introduce unwanted function, or subvert a covered system.

I think any LLM is covered by that, but specifically for Anthropic,

>Recent research has uncovered several critical vulnerabilities, including the "Claudy Day" attack chain which allows silent data exfiltration through conversation history, and a zero-click XSS prompt injection in the Chrome extension that enabled attackers to inject prompts without user interaction until a patch was released in February 2026.

What is obvious to me however is the timing. This Trump pants-shitting happened just before the Iran invasion. You can just imagine it. Trump wants to send fully autonomous bots into Iran to destroy the non-existent nuclear program. Anthropic leadership tries to make a moral stand saying innocent civilians could die. Trump doesn't care because he wants zero US military casualties even if it means a school full of Iranian children is bombed and everyone is killed. And then we get exactly that plus a forever war.

And obviously, the judge is out of her lane too... since, you know, the rule basically can apply to any AI agent because they're just as likely to do what you ask as they are to delete all your emails without even apologizing for it.