Anthropic Rejects Pentagon's Demand to Remove Claude's Restrictions

Anthropic is standing firm, having rejected the Pentagon's ultimatum. CEO Dario Amodei issued a clear statement yesterday afternoon saying that the company cannot, in good conscience, accept the Department of Defense's request to remove safeguards from Claude.

The government wants to use the model without any restrictions, for any legal purpose, including:

• Mass surveillance of American citizens — which is illegal in the US, but they want total flexibility
• Autonomous weapons that select targets without genuine human oversight

Amodei made it clear that Claude is already operational within classified networks and national labs, so collaboration exists; however, on these specific issues, they won't budge. He argues that frontier models like Claude are still not reliable enough to make life-or-death decisions without strong guardrails, without a human in the loop to check for hallucinations or grave errors. The company even offered to work together on research to improve reliability, but the Pentagon wasn't interested.

On the other side, Defense Secretary Pete Hegseth has set a deadline: by 5:01 p.m. ET today (22:01 UTC), Anthropic must comply or face severe consequences:

• Losing the $200 million contract
• Being labeled a supply chain risk, a designation typically reserved for companies linked to adversarial countries
• Invoking the Defense Production Act to force access

Spokesperson Sean Parnell posted on X that they will not allow any company to dictate how they make operational decisions and reiterated the deadline.

Anthropic responded that the threats are contradictory: on one hand, they call it a security risk, on the other, they say Claude is essential for defense. More than 200 employees from Google and OpenAI signed an open letter supporting their position, arguing that these safeguards protect democratic values.

Ultimately, this highlights the clash between real AI ethics and military pressure to accelerate everything without brakes. For those using Claude or other LLMs daily, it means the guardrails that prevent dangerous, repetitive, or off-the-wall responses might remain intact, at least for now. The clock is ticking, and any response from the Pentagon tonight is likely to intensify the debate.