The Maven rebellion is back, and this time the engineers know they're already too late to stop it.

The Summary

The Signal

This is Maven 2.0, but the script has flipped. In 2018, Google employees revolted against Project Maven and won. The Pentagon contract died. Google pledged AI ethics principles. Engineers celebrated. The company learned to stay away from defense work that made headlines.

But here's what changed: the frontier models got too good to keep off the battlefield. Google didn't need to seek out defense contracts. The Department of Defense came shopping with a credit card, and somewhere in the last two years, Google's leadership decided the ethical red lines were negotiable if the work stayed classified.

"The only way to guarantee that Google does not become associated with such harms is to reject any classified workloads."

The employees' argument is tight: you can't have "responsible AI" if you can't see what your AI is doing. Classified deployments mean zero visibility into whether models are being used for targeting decisions, surveillance infrastructure, or autonomous weapons systems. The engineers aren't arguing against all defense work. They're arguing against the impossibility of oversight in secret.

Key facts the letter exposes:

  • Google's AI ethics principles from 2018 are now in direct conflict with its business reality
  • Senior technical leadership (principals, directors, VPs) are willing to put their names on internal dissent
  • The timing suggests classified Pentagon deals are already happening or imminent

What makes this different from Maven is scale and stakes. In 2018, Google was partnering on narrow computer vision tasks. Today, we're talking about frontier models, Gemini-class systems that can reason, plan, and generate strategic options. The blast radius of deployment is exponentially larger.

And the Anthropic detail matters more than it looks. If Anthropic is in court fighting the Pentagon, it means the DoD is aggressively pursuing multiple foundation model providers. This isn't one rogue contract. It's a procurement strategy. Google's employees see the pattern: their models will end up in classified systems whether leadership admits it publicly or not.

Structural tension no letter can solve:

  • Google needs compute-heavy customers to justify its infrastructure spend
  • The Pentagon has effectively infinite budget for AI and writes checks faster than enterprise sales cycles
  • Classified work pays premium rates and doesn't require the messy public accountability of commercial products

The 600 signatures signal cultural fracture, not just policy disagreement. When principals and directors sign dissent letters, it means management has lost the room. These aren't junior engineers making noise. These are people who've been at Google long enough to remember when "Don't be evil" wasn't ironic.

The Implication

If you're building AI, the choice is coming for you too. The defense market is the most lucrative customer base in the AI economy right now, and it doesn't care about your principles page. You'll either take the contracts and manage internal revolt, or you'll lose ground to competitors who will.

For Google employees still inside: the letter is a paper trail. It's not about changing Pichai's mind. It's about having evidence on record that technical staff raised alarms before the inevitable incident. When something goes wrong with a classified deployment, this letter becomes Exhibit A in the "we told you so" file.

Watch what Google does in the next 30 days. If leadership ignores the letter, expect departures. If they engage, expect watered-down "transparency reports" that reveal nothing about classified work. Either way, the frontier models are going to war. The only question is whether the people who built them stay in the room.

Sources

The Verge AI