OpenAI and Google staff back Anthropic's lawsuit against the Pentagon

Silicon Valley's fiercest rivals just broke rank to back each other against the government.

On Monday, Anthropic filed two lawsuits against the Pentagon over being labeled a supply chain risk. Hours later, an amicus brief landed in court signed by nearly 40 engineers and researchers from OpenAI and Google DeepMind, including Google chief scientist and Gemini lead Jeff Dean.

The standoff began after Anthropic refused to budge on two conditions for military use: no domestic mass surveillance on Americans, and no fully autonomous lethal weapons that kill without human intervention. The Department of Defense wanted free rein for any lawful use. When talks collapsed, the Trump administration retaliated with a designation usually reserved for foreign adversaries.

The blacklist crossfire

The supply chain risk label carries devastating reach. It does not just bar Anthropic from military deals. It effectively blacklists any third-party defense contractor using Claude, forcing them to strip the model out of their tech stacks if they want to keep Pentagon money.

Yet Claude is already woven into classified intelligence operations. In an ironic twist, Defense Secretary Pete Hegseth slapped the label on Anthropic just hours before the U.S. military reportedly used Claude in the strike that killed Iranian leader Ayatollah Ali Khamenei.

Adding insult to injury, the DOD turned around and signed a deal with OpenAI immediately after blacklisting Anthropic. That triggered fierce blowback from OpenAI's own workforce, many of whom joined the court filing.

Unpacking the red lines

The researchers made clear they were signing as individual professionals who understand exactly what these models can and cannot do. If the military disliked Anthropic's terms, they argued, it could have simply walked away and hired a rival instead of weaponizing an arbitrary designation to punish dissent.

The brief laid out why Anthropic's red lines matter. Massive piles of surveillance footage, location traces, and credit card data already exist, but they are trapped in separate silos. The danger is that an AI layer can fuse those fragments in real time, tracking hundreds of millions of people at once.

When it comes to autonomous weapons, the engineers warned that frontier models hallucinate and stumble in unfamiliar environments. Algorithms cannot weigh collateral damage the way a human can, making human oversight essential before launching lethal munitions.

Why it matters

This legal fight is drawing the ultimate battle line between AI safety principles and state power.

For enterprise customers and builders, the Pentagon's retaliation is a chilling precedent. If a domestic vendor can be treated like a hostile foreign power just for demanding contract guardrails, and if downstream partners face blacklisting for integrating that vendor's API, the risks of corporate compliance suddenly become existential.

We think the cross-lab solidarity from OpenAI and Google employees is remarkable. They recognize that if the military successfully bullies one lab into dropping safety standards through contract leverage, the entire industry enters a race to the bottom where everyone loses autonomy.

The hard reality, however, is that government procurement operates on realpolitik. The DOD proved it can immediately pivot to whoever signs on the dotted line. Anthropic took an enormous financial gamble by drawing these boundaries, and while this lawsuit might expose executive overreach, keeping the moral high ground in Washington could cost Claude its most lucrative enterprise footprint.