Pentagon demanded Anthropic enable weapons
Unsealed court emails show the Pentagon demanded Anthropic remove guardrails on autonomous weapons and domestic surveillance; Amodei refused.
Court documents unsealed July 2 reveal emails between Emil Michael, the Pentagon’s Under Secretary for Research and Engineering, and Anthropic CEO Dario Amodei. Michael demanded that Anthropic remove two categories of guardrails from Claude: restrictions on autonomous lethal weapons decisions and restrictions on domestic mass-surveillance applications.
Amodei refused. Financial disclosures filed at the time show Michael held $2 million to $10 million in Perplexity stock, a direct Anthropic competitor, and had sold xAI shares generating roughly $24 million in proceeds (reported range: $5 million to $25 million) while serving as the Pentagon official making AI vendor decisions.
In or around March 2026, the Pentagon designated Anthropic a “supply chain risk,” blocking it from certain government contracts. Anthropic challenged the designation in court. Amodei’s public statement:
“we do not believe this action is legally sound, and we see no choice but to challenge it in court.”
That document release on July 2 is part of the ongoing litigation.
What it means for you
The officials making AI procurement decisions for government clients are sometimes holding stakes in rival AI companies. The disclosure rules designed to catch that are still being built for this industry.
For operators: Anthropic’s Fable 5 guardrails are deliberately visible partly so third parties can see what the company refused to remove. That is a specific choice, not a default. Alongside US export restrictions on top Claude models, this case shows government policy and safety commitments in direct conflict, with AI vendors being pressed to choose.