Over 600 Google Employees Tell Sundar Pichai: No Military AI for the Pentagon

Google employees are at it again, but this time the stakes feel higher.

More than 600 staffers — including over 20 principals, directors, and vice presidents, many from DeepMind — signed a letter to CEO Sundar Pichai telling him to block the Pentagon from using Google’s AI models for classified military work. The Washington Post broke the story, and it’s a reminder that the company’s internal conflict over military contracts never really went away.

The letter doesn’t mince words: “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.”

That’s a direct shot at the kind of black-box deployment where Google hands over its AI, loses visibility, and ends up complicit in whatever the military does with it. The employees aren’t asking for oversight or transparency clauses. They want a hard no.

This isn’t the first time Google’s AI ethics have collided with Pentagon dollars. Remember Project Maven back in 2018? That was the drone imagery analysis contract that sparked massive internal revolt, culminating in Google deciding not to renew it and publishing a set of AI principles that explicitly banned weapon systems. But those principles had loopholes, and it looks like the Pentagon found one.

What’s new here is the classified angle. The employees are worried that once the work is behind security clearances and sealed doors, there’s no way to audit whether the AI is being used for targeting, surveillance, or something even uglier. They’re essentially saying: if you can’t see what’s happening, assume the worst.

Meanwhile, Anthropic — the “safety-first” AI company founded by ex-OpenAI folks — is currently in a legal battle with the Pentagon over similar issues. That’s ironic, given how much Anthropic markets itself as the responsible alternative. Turns out, when the money is big enough and the mission is “national security,” ethical lines get fuzzy fast.

I’ve been watching this tension play out across the industry for years. Every major AI lab has some version of this fight inside its walls. The difference with Google is scale. DeepMind alone has some of the brightest minds in the field, and if they’re organizing, it’s not just a PR headache — it’s a brain drain risk. Nobody wants to be the person who built the AI that helped target a school.

Pichai hasn’t responded publicly yet, and Google’s official stance is probably being workshopped by a dozen lawyers as we speak. But the clock is ticking. The Pentagon isn’t going to wait, and there are plenty of other vendors — Palantir, Amazon, Microsoft — lining up to take Google’s seat at the table.

What I’d really like to see is a clear, public audit of what Google’s AI principles actually mean in practice. Right now, they read like a press release: “we will not design AI for weapons.” But “design” is doing a lot of heavy lifting. What about deploying existing models for military logistics? What about using AI to analyze satellite imagery that feeds into targeting decisions? That’s not “designing” a weapon, but it’s still enabling one.

The employees are right to be skeptical. Once you hand over the keys to a classified system, you don’t get to ask questions. And for a company that built its brand on “Don’t Be Evil,” that’s a hell of a thing to sign off on.