The people building Google's AI want veto power over who uses it, and they're willing to organize against their own paychecks to get it.
The Summary
- Around 600 Google employees, including DeepMind and Cloud division staff, sent a letter to CEO Sundar Pichai demanding the company reject classified Pentagon AI contracts.
- The letter responds to reports that Google and the Pentagon are negotiating use of Gemini AI in classified settings, raising concerns about autonomous weapons and surveillance.
- Workers argue their proximity to the technology creates a responsibility to prevent its "most unethical and dangerous uses," and that classified work would happen "without our knowledge or the power to stop them."
The Signal
This is the second act of a play Google already performed. In 2018, employee protests killed Project Maven, a Pentagon contract using Google's AI for drone video analysis. That uprising forced Google to adopt AI Principles promising not to build weapons systems. Now 600 workers are testing whether those principles mean anything when billions in defense contracts are on the table.
The timing matters. Every major AI lab is racing toward government partnerships. OpenAI just hired its first Chief of Defense. Anthropic is pitching Claude to intelligence agencies. Scale AI built its entire business on defense data labeling. Google's employees are drawing a line while competitors sprint past it.
"Currently, the only way to guarantee that Google does not become associated with such harms is to reject any classified workloads."
The letter's core argument is structural, not philosophical. Workers claim that once AI systems enter classified environments, they lose visibility into how they're used. No oversight. No ability to object. No way to know if Gemini is analyzing satellite imagery or targeting packages. The classified barrier makes Google's AI Principles unenforceable promises.
What makes this different from 2018:
- The signatories include DeepMind researchers, Google's crown jewel AI team
- The ask is categorical: reject ALL classified work, not just weapons projects
- Google's Cloud business now depends on enterprise and government revenue in ways it didn't six years ago
Jane Chung of Justice Speaks, representing the workers, confirmed Google hasn't responded to the letter. That silence is its own signal. In 2018, Google's leadership could afford to pick employee morale over a single contract. In 2026, with Microsoft and Amazon Web Services embedded in defense infrastructure, the cost of saying no is measured in market share.
The Implication
Watch what happens next. If Pichai caves again, Google's AI business becomes the only major lab explicitly barred from classified government work. That's a competitive anchor in a market where federal contracts are table stakes. If he ignores the letter, he's betting 600 employees will stay despite public opposition, or that he can replace them faster than they can organize.
For anyone building AI companies, this is the canary. Your early team's values will eventually collide with your growth strategy. The question isn't whether that collision happens. It's whether you've built a company culture that can survive it, or whether your best people organize against you when the checks get big enough.