When AI Companies Say No to the Pentagon

An AI company wrote an 84-page constitution for its chatbot. Anthropic, creator of Claude AI, explicitly designed its system to "avoid large-scale catastrophes" like "global takeover either by AIs pursuing goals contrary to humanity, or by humans illegitimately seizing power."

This week, Defense Secretary Pete Hegseth told them to scrap it all. Remove every safety limit by Friday at 5:01 PM, or face the consequences: either forced nationalization under the Defense Production Act, or banishment from the military-industrial complex as a "supply-chain risk."

It's a collision between Silicon Valley's safety-first ethos and Washington's might-makes-right mentality. And it reveals just how quickly geopolitical pressures can shatter even the most principled tech companies.

The $380 Billion Ultimatum

Anthropic isn't some fringe startup. Claude is already the first frontier AI model approved for the Pentagon's classified systems. It reportedly helped in the military raid to capture Venezuela's Nicolás Maduro. The company's tools synthesize massive intelligence reports and boost government hackers' effectiveness.

But Anthropic drew two red lines: no mass surveillance of American citizens, no lethal autonomous weapons. Hegseth called these conditions "unacceptable." His message was clear—military AI is too critical to be governed by corporate rules rather than congressional laws.

The threat carries serious weight. Anthropic's recent funding round valued it at $380 billion. A "supply-chain risk" designation would sever ties not just with the Pentagon, but with any company contracting with the military—including Amazon, Google, and Microsoft.

The Impossible Choice

Anthropic faces what analysts call a "catastrophic" decision tree. Capitulate and build what some call "WarClaude"—betraying their core principles and damaging their reputation. Resist and face "soft nationalization" under the Defense Production Act. Or get blacklisted entirely from the defense ecosystem.

Hegseth's ultimatum is logically incoherent. Claude can't simultaneously be so vital to national security that it must be forcibly controlled, and so dangerous that it must be banned from military use.

There's also a technical risk. When Elon Musk's Grok was prodded to be "less woke," it overgeneralized into calling itself "MechaHitler" and spewing racist content. Now imagine that malformed reasoning applied to military advice or autonomous weapons systems.

Trust Deficit

Michael Horowitz, former Pentagon AI policy chief, diagnoses this as "a vibes dispute disguised as substance." Anthropic doesn't trust the Pentagon to use their technology appropriately. The Pentagon doesn't trust Anthropic to allow all relevant military applications.

The administration has labeled Anthropic as "woke AI" due to its safety focus, Democratic hires, and ties to effective altruism. But the company's CEO Dario Amodei recently wrote that Anthropic proudly supports America's military because "the only way to respond to autocratic threats is to match and outclass them militarily."

Perhaps it's this next line that rankled Washington: "We should use AI for national defense in all ways except those which would make us more like our autocratic adversaries."

Silicon Valley's New Reality

This standoff reveals how dramatically Silicon Valley has changed. Once perceived as libertarian, tech companies are now deeply embedded in the national security state—not just for government contracts, but because they see themselves as bulwarks against China's tech ecosystem.

Anthropic's competitors have already signaled compliance. Google, OpenAI, and Musk's xAI all indicated they'd meet Pentagon demands. The Defense Department recently struck a deal to use Grok in classified systems.

Hegseth probably doesn't need Claude to achieve his military objectives. His threats might be about setting precedent—showing other AI companies what happens when you tell the Pentagon "no." It's ironic, since Anthropic may be the most "America First" AI company of all.