More than 600 Google employees, many from the elite DeepMind lab, fired off a letter to CEO Sundar Pichai this week. They demand he block any Pentagon deal allowing classified use of Google’s AI models. Signers include over 20 principals, directors, and vice presidents. Their message is blunt: reject classified workloads outright. Otherwise, harms could unfold without Google’s knowledge—or power to intervene.
The pushback comes hard on the heels of a report from The Information detailing Google’s talks with the Defense Department. Those discussions aim to deploy Gemini, Google’s flagship model, in classified settings. This follows the Pentagon’s bitter split with Anthropic, labeled a supply-chain risk after refusing unrestricted access to its Claude models.
“As people working on AI, we know that these systems can centralize power and that they do make mistakes,” the Googlers wrote. “We feel that our proximity to this technology creates a responsibility to highlight and prevent its most unethical and dangerous uses.” They spotlight risks like lethal autonomous weapons and mass surveillance—precisely the red lines that sank Anthropic’s deal.
History repeats. Back in 2018, over 3,000 Google workers protested Project Maven, a drone-imaging AI contract. Google canceled it amid the uproar. But times have changed. The company now holds non-classified DoD pacts via genAI.mil and has quietly expanded military ties. Employees see classified work as a bridge too far. No oversight. No recourse. Just potential complicity.
And yet. Google filed an amicus brief backing Anthropic’s lawsuit against the Pentagon earlier this year, alongside OpenAI staff. That showed solidarity on safety bounds. Now, with the Pentagon shopping elsewhere, Google’s foot soldiers revolt while leadership eyes the prize. A DeepMind engineer put it starkly to organizers: classified means opacity, where tools might target civilians unchecked.
Pichai hasn’t responded publicly. Silence speaks. The letter, coordinated by DeepMind and Cloud staff, hit his desk Monday. By Tuesday, it leaked to The Washington Post and The Verge. CBS News called it a direct shot at ongoing DoD negotiations. Business Insider noted the signers’ seniority, underscoring internal clout.
But 600 out of 182,000 is less than 0.4%. Enough to make noise, perhaps not policy. Google killed Maven, sure—then signed a $1.2 billion DoD cloud deal in 2021 under another name. Playbook intact: protest, PR spin, proceed.
Anthropic’s fate looms large. Designated a national security risk, it’s in court. Google and OpenAI grabbed contracts in the void, despite their own workers’ petitions. OpenAI staff echoed the call for restrictions. Yet Sam Altman inked a deal allowing “all lawful uses,” sidestepping the fight Anthropic picked.
So what’s at stake? Control. Employees argue Google can’t enforce bans in black-box classified nets. Pentagon wants full latitude—surveillance, targeting, whatever “lawful.” Google proposes clauses against autonomous killers or domestic spying. Workers say that’s toothless. Once deployed, models evolve in shadows.
This rift exposes AI’s fault lines. Tech giants chase defense dollars amid China rivalry. Revenue beckons: billions in cloud, chips, models. But talent—DeepMind’s brainiacs—baulks. Sofia Liguori, a UK-based DeepMind AI engineer, signed on. She faults Google for dodging worker input on red lines.
Broader wars rage too. Google pumps billions into Anthropic—$3 billion for 14% stake, now paper worth over $100 billion per X chatter. Yet DeepMind engineers crave Claude for coding; access fights brew internally. Rivals poach. Protests risk exodus.
Pentagon presses on. OpenAI’s in. Palantir thrives. Google? Pichai must weigh ethics, stock, staff. Ignore the letter, and morale frays. Heed it fully, and rivals feast. Partial yield—like Maven exit—might split the difference. But classified AI isn’t going away. It’s the new arms race.
Employees aren’t naive. They cite Maven’s ghost. Demand transparency. Pichai’s call could redefine Google’s soul—or just its contracts. Watch the silence break.
WebProNews is an iEntry Publication