600 Google Employees Tell Sundar Pichai: No Classified Military AI

Six hundred Google employees just signed a letter to CEO Sundar Pichai, and the message is blunt: don’t let the Pentagon use our AI for classified work.

The Washington Post broke the story. The letter’s organizers say signers include over 20 principals, directors, and VPs—many from DeepMind. That’s not a handful of junior engineers grumbling on Slack. These are people who actually build the stuff.

The core argument is simple but damning: “The only way to guarantee that Google does not become associated with such harms is to reject any classified workloads. Otherwise, such uses may occur without our knowledge or the power to stop them.”

Translation: once the black box closes, nobody inside Google gets to audit what the military does with their models. And that’s exactly what worries them.

This isn’t Google’s first rodeo on military AI. Remember Project Maven in 2018? That internal revolt forced Google to back away from drone imagery analysis contracts. The company later published AI principles explicitly banning “weapons or other technologies whose principal purpose or implementation is to cause or directly facilitate injury to people.”

But principles are cheap. The question is whether Google’s leadership actually enforces them when the Pentagon comes calling with a fat contract.

Meanwhile, Anthropic is already in a legal fight with the Pentagon over similar issues. That case will set some precedent, but it’s moving slowly. Employees at Google aren’t waiting for courts to figure it out.

I’ve seen this pattern before. Tech companies love publishing ethical AI frameworks during good times. But when defense dollars show up, those frameworks suddenly need “careful interpretation.” The employees know this. That’s why they’re pushing for an outright ban on classified work—no loopholes, no carve-outs for “defensive” systems.

The real tension here isn’t between employees and management. It’s between the nature of AI research and the nature of military contracting. AI models are general-purpose tools. You can fine-tune a language model for medical diagnosis or for targeting systems. The same architecture, different data. That generality makes it nearly impossible to say “our AI won’t be used for X” once it leaves your control.

Google’s response so far has been predictable: we follow the law, we have ethical guidelines, we’ll review carefully. But 600 signatures suggest that’s not going to cut it this time.

I’ll be watching how Pichai handles this. If he caves to Pentagon pressure, expect more resignations from DeepMind. If he backs the employees, expect a very awkward phone call from the Department of Defense.

Either way, the cat’s out of the bag. Employees at the world’s leading AI lab just publicly declared they don’t trust their own company’s leadership to keep military AI in check. That’s not a PR problem. That’s a structural crisis.