Pentagon gives Anthropic an ultimatum: drop Claude safeguards or face blacklisting

The Pentagon just gave an American AI lab an ultimatum that sounds like something out of a techno-thriller: strip away your safety guardrails, or get treated like an enemy foreign threat.

According to Axios, the U.S. Department of Defense has reached out to major defense contractors Boeing and Lockheed Martin to evaluate their reliance on Anthropic's Claude. The assessment is the first concrete step toward potentially designating Anthropic a "supply chain risk" — a severe blacklist penalty traditionally reserved for adversarial foreign firms like Huawei.

Defense Secretary Pete Hegseth delivered a strict deadline to Anthropic CEO Dario Amodei: Friday at 5:01 pm. The demand is simple and blunt. Anthropic must remove Claude's military safeguards, or face severe consequences.

Remove the guardrails or face the blacklist

At the heart of the standoff are two clear ethical lines Amodei has refused to cross: allowing Claude to be used in autonomous weapons without a human in the loop, and using the AI for bulk surveillance of American citizens.

Axios reports that Hegseth gave Anthropic three choices. They can agree to drop the limits, see their $200 million defense contract terminated under a supply chain risk designation, or potentially be forced to comply under the sweeping authorities of the Defense Production Act.

Claude is currently the only frontier model running directly inside U.S. military classified systems, but the Department of Defense is already signaling that nobody is indispensable. xAI recently inked a military deal for Grok after agreeing to terms covering "all lawful purposes." Meanwhile, the Pentagon is reportedly fast-tracking both OpenAI and Google for classified clearance, giving Hegseth immediate alternative leverage if Anthropic decides to walk away.

Why it matters

We see this as the most alarming pressure test yet between private AI safety commitments and state power.

For enterprise customers and developers, Anthropic has built its entire brand around safety and predictable alignment. If the Pentagon can simply wield wartime legislation and national security blacklists to force a domestic company into stripping safeguards against mass surveillance and killer robots, voluntary industry guardrails effectively cease to exist.

From a military standpoint, Hegseth wants unencumbered capability, not moral lectures from software code. But strong-arming a premier domestic lab sets a toxic precedent across Silicon Valley. It penalizes ethical friction while actively rewarding competitors willing to sign off on "all lawful purposes."

If national security threats are all it takes to force frontier AI models into autonomous warfare, the scary reality is that the boundary between responsible deployment and dystopian overreach is no longer being drawn by the people building the technology.