Anthropic refuses military surveillance demands, faces blacklist threat

When government negotiations go south, parties usually agree to disagree and walk away. But this week, a standard commercial impasse escalated into an unprecedented showdown between the Pentagon and one of America’s leading AI labs.

Secretary of War Pete Hegseth announced on X that he is directing the department to designate Anthropic as a supply chain risk. President Donald Trump quickly followed up by ordering all federal agencies to phase out Anthropic tools after a six-month transition period.

Hegseth went a step further, declaring that effective immediately, no contractor, supplier, or partner doing business with the military may conduct any commercial activity with Anthropic.

The two red lines that broke the deal

The dispute centers on two specific carve-outs Anthropic demanded before letting Claude power military systems: no mass domestic surveillance of American citizens, and no fully autonomous weapons capable of targeting and firing without a human in the loop.

Anthropic argued that current frontier models are simply not reliable enough to make life-or-death battlefield decisions safely, and that mass surveillance violates fundamental constitutional rights. The lab noted that Claude has supported classified military networks since June 2024 without these limits disrupting a single mission.

The Pentagon reportedly stated it had no plans to do either, but refused to let a commercial vendor dictate operational boundaries to the military.

Anthropic pushed back hard, calling the proposed designation legally unsound and unprecedented for an American company. The firm clarified that under federal statute, supply chain risk designations only apply to work performed under specific defense contracts, meaning commercial users, API customers, and even military contractors working on non-defense projects remain completely unaffected.

Tech workers push back as OpenAI steps in

The administration’s aggressive stance sparked immediate backlash across Silicon Valley. Hundreds of tech workers from firms including OpenAI, Slack, IBM, Cursor, and Salesforce Ventures signed an open letter urging the Pentagon and Congress to withdraw the blacklist.

The letter warned that weaponizing national security powers against a domestic company for negotiating terms sets a terrifying precedent: comply with every government demand or face economic retaliation.

Yet the geopolitical vacuum didn't stay open for long. Minutes after the White House attacked Anthropic, OpenAI announced its own deal to deploy models into the Pentagon’s classified networks.

While OpenAI CEO Sam Altman claimed his company shares the exact same ethical red lines, OpenAI researcher Boaz Barak publicly called on the industry to treat government abuse and mass surveillance as catastrophic risks on par with bioweapons and cyber threats.

Why it matters

For existing Claude users and enterprise developers, nothing changes today. Anthropic’s consumer apps and commercial APIs are operating as normal, and Hegseth’s threat to ban private contractors from using Claude across their entire commercial stack lacks clear legal authority.

For the broader AI ecosystem, however, the stakes could not be higher. We think the Pentagon’s move is a dangerous overreach that weaponizes procurement regulations to crush corporate dissent. If an American vendor can be labeled a national security hazard simply for refusing to build surveillance and killer robots, ethical guardrails across the entire industry become meaningless overnight.

Anthropic deserves credit for holding its ground at massive financial cost, but it now faces an uphill legal battle against executive power. Meanwhile, OpenAI’s opportunistic scramble to scoop up the military contract looks tactically sharp, but it puts Sam Altman on the hot seat. The moment defense officials ask OpenAI to drop those identical red lines, we will see whether its principles hold up any better under fire.