Defense Secretary Pete Hegseth has warned Anthropic to drop its limits on AI use or lose a $200 million government contract. The clash happened in a meeting Tuesday between Hegseth and Anthropic CEO Dario Amodei. It's about Claude, the company's chatbot cleared for top-secret military work. Hegseth wants no restrictions on how the Pentagon uses it. Anthropic won't budge on bans against mass surveillance or killer robots. Friday's the deadline. Cut ties. Or face bigger trouble.
Key Takeaways
- Pete Hegseth gave Anthropic until Friday evening to allow unrestricted military access to Claude AI.
- Claude is the only AI model the Pentagon uses in classified operations, including a recent Venezuela raid.
- Anthropic faces contract cancellation, supply chain risk label, or Defense Production Act force.
- The company agrees to tweak rules for the Pentagon but draws lines at surveillance and autonomous weapons.
Background
Pete Hegseth took over as Defense Secretary early in the Trump administration. He's pushed hard for U.S. leads in military tech. AI tops that list. The Pentagon needs fast, powerful tools for cyber ops, planning, and more. Anthropic built Claude as a safe AI. It follows strict rules. No helping with spying on Americans. No building weapons that kill on their own. That's drawn fire from officials who see it as too cautious. Or "woke," as some call it.
Claude got special approval for secret work. No other AI matches it there. The military uses it through partners like Palantir. That firm helped in a January raid on Venezuela's Nicolas Maduro. U.S. forces grabbed him in Caracas. Claude aided planning. Reports say Anthropic talked to Palantir after. They asked about use in that op. But Anthropic says it was just routine chat on policy. No complaints. No blocks on field work.
Tensions built fast. By last week, Pentagon brass eyed cuts. They see Anthropic's rules as roadblocks. Hegseth agrees. He won't let any firm tell the military what it can do. This fight tests how far safety limits go in defense deals. Other AI firms watch close. Like AMD and Meta's huge AI chip pact. Deals like that show the rush for power. But Claude stands alone in secrets.
Anthropic raised $30 billion this year. Some investors passed. Firms tied to Trump skipped it over the company's safety push. That's the backdrop. Ethics versus speed. Now it boils over.
Key Details
Hegseth met Amodei Tuesday. High stakes. He brought top aides. Deputy Secretary Fein. Under Secretary for Research Emil Michael. Chief spokesperson Sean Parnell. General counsel Earl Matthews. Serious crew. The talk stayed calm. No yelling. Hegseth even praised Claude's smarts. But firm words flew.
"Give us full access by Friday night," Hegseth said, per sources. Or else. Option one: Kill the $200 million contract. That's for Claude in government work. Option two: Label Anthropic a "supply risk." That bars it from defense chains. Forces others to ditch Claude too. Option three: Invoke the Defense Production Act. Rare move. President orders firms to meet defense needs. Used in COVID for masks and shots. Here, it'd force Anthropic to tweak Claude. Strip safeguards.
The Friday Deadline
Deadline hits Friday evening. Short fuse. Anthropic says it's open to changes. Just for Pentagon needs. But core lines hold. No mass U.S. spying. No auto-killers. Amodei denied Pentagon claims. He says Anthropic never griped to Palantir on the Maduro raid. Just standard policy talk. No ops slowed. No troops hurt.
"During the dialogue, Dario expressed gratitude for the Department's efforts and thanked the Secretary for his service. We continued our constructive discussions regarding our usage policy to ensure Anthropic can keep supporting the government's national security mission in alignment with the capabilities of our models." – Anthropic spokesperson
Pentagon insiders split on tone. One called it "not warm and fuzzy." Another said cordial. Hegseth clear: No company dictates terms.
Claude does more than raids. Cyber attacks. Admin tasks. It's top for offensive cyber, sources say. Rivals lag. Elon Musk's xAI just signed for classified Grok use. But unproven. OpenAI and Google models work unclassified. Pentagon pushes to upgrade them. Google Gemini might step up. If it drops some rules.
What This Means
Lose Claude? Big gap. No quick fix. Pentagon scrambles. Other AIs need clearance. That takes time. Money. Meanwhile, foes advance. China. Russia. They don't pause for ethics.
Anthropic risks billions. Government cash flows big. But sticking to principles sets tone. Shows safety matters. Or hurts. Investors note. 1789 Capital, with Trump Jr. links, already bailed on funding.
Broader shift. Trump team wants AI speed. No brakes. States push rules. California demands safety reports. EU has its AI Act. India talks safety too. U.S. lead slips if it fights allies.
Anthropic could sue. Argue Claude's custom. Not off-shelf goods. Defense Production Act might not fit. Courts decide. Slows everything.
Military feels pinch first. Field teams rely on Claude. Cyber units. Planners. Switch hurts ops. Sources admit: They need it bad. Quality unmatched.
This feud spotlights the split. Safety firms versus full-throttle needs. Pentagon weighs cut versus keep. Friday looms.
And firms like those in Google's AI paths push boundaries too. But classified? Claude rules still.
Frequently Asked Questions
What is Claude?
Claude is Anthropic's AI chatbot. It's built with safety rules. The Pentagon uses it in secret ops because it's the only one cleared there.
Why the fight now?
It stems from Claude's use in a Venezuela raid. Pentagon says Anthropic complained. Company denies. Wants no surveillance or autonomous weapons.
Can the Pentagon force changes?
Yes, via Defense Production Act. Rare. Or cut ties. Label supply risk. But no replacement ready.
Frequently Asked Questions
What is Claude?
Claude is Anthropic’s AI chatbot. It’s built with safety rules. The Pentagon uses it in secret ops because it’s the only one cleared there.
Why the fight now?
It stems from Claude’s use in a Venezuela raid. Pentagon says Anthropic complained. Company denies. Wants no surveillance or autonomous weapons.
Can the Pentagon force changes?
Yes, via Defense Production Act. Rare. Or cut ties. Label supply risk. But no replacement ready.
