The letter was 600 names long.

It went up Monday morning. Senior DeepMind researchers. Twenty-plus VPs and directors. Engineers from across Google's AI organization. The ask was specific: do not sign the classified Gemini contract with the Department of Defense. Do not give the Pentagon access to Google's frontier models for use on air-gapped networks where the company cannot monitor what its own technology is doing. "Making the wrong call right now," the letter told CEO Sundar Pichai, "would cause irreparable damage to Google's reputation, business and role in the world."

Pichai signed the deal Tuesday.

The Information broke the news first. Reuters, The Guardian, CBS, 9to5Google and The Decoder all matched it within hours. The agreement gives the Pentagon access to Google's Gemini models for "any lawful government purpose," the boilerplate phrase Anthropic and OpenAI used in their own military contracts and the phrase that means classified work, including defense applications, is on the table.

Less than 24 hours from open letter to signed contract. That has to be a record.

What Just Happened

Google's previous public position was the 2018 Project Maven retreat. After thousands of employees protested the company's drone-imagery work for the Pentagon, leadership pulled out of the contract and published an AI principles document with a specific ban on weapons applications. That ban was quietly removed from the principles in early 2025. The company has since accepted a slice of the $9 billion Joint Warfighting Cloud Capability contract. Gemini is now deployed to roughly 3 million Pentagon personnel on unclassified networks. The new deal extends that into classified work.

The 600-name letter, per Bloomberg's earlier reporting, came from inside DeepMind specifically. That matters. DeepMind is where Google's most expensive and most mobile AI talent lives. It is also the lab Google paid $40 billion in 2024 to keep separate from the rest of the company, then folded back in when AI scaling forced consolidation. The signatories include the kind of senior researcher who can leave on a Wednesday and have a billion-dollar seed round announced by Friday.

They have a track record on that. David Silver, the AlphaGo researcher, left DeepMind in March and announced Ineffable Intelligence with $1.1 billion in seed funding two weeks ago. Koray Kavukcuoglu, formerly DeepMind's VP of research, raised $1 billion for AMI Labs the same month. Two of the most important AI researchers in the company walked out the door this quarter. The classified Pentagon deal is the third reason in 60 days for the next one to follow.

Anthropic Already Lost This Argument. Two Months Ago.

This is the part the news write-ups have not connected.

In late February, the Pentagon dropped Anthropic from a vendor evaluation after the company asked for the same monitoring carve-outs the Google letter is now demanding. OpenAI signed the contract within hours, with three voluntary red lines on offensive use. Anthropic walked away from a roughly $200 million annual revenue line because its own safety framework wouldn't permit blind classified deployment. Two months later, Google's own employees are demanding Google take the Anthropic position. Google took the OpenAI position.

So the scoreboard is now: OpenAI yes, Google yes, Anthropic no. Three of the four largest American AI labs have now signed Pentagon classified contracts. The one company that said no is the one that just took a $40 billion infrastructure investment from Google and is preparing to file for an October IPO, where its safety-first posture is the entire pitch to public investors.

There Is No Coherent Position Left

Here is what the letter actually says, beyond the Pichai-shaped headline.

Frontier AI labs cannot be both the moral authority on AI safety and the prime contractor on classified military AI. They have spent five years telling regulators, journalists, governments, customers and their own employees that the technology is so powerful it might end the world. Then they sign contracts that let three-letter agencies use it on air-gapped networks no third party can audit. The reconciliation, when pressed, is some version of: better us than them.

That argument has a window. The window closed in February when Anthropic refused. It closes harder this week, because the moment one of the big labs signs a classified contract over the loud, public, internal objection of 600 of its most senior researchers, the position becomes unsustainable. Either AI is a frontier safety problem and these contracts are reckless, or it is a mature commercial product like every other commercial product and the safety theater is over.

The 600 signatories are about to find out which Google believes.

Some of them will leave. The talent market for senior AI researchers is the tightest it has been in the company's history. AMI Labs, Ineffable, Recursive Superintelligence and Humans& are all hiring with billion-dollar war chests. A signed Pentagon deal that the labs' own staff publicly opposed is a recruiting brochure for every one of them.

The other 599 are going to keep building Gemini.

Pichai made a calculation: the revenue is bigger than the protest, the talent is replaceable at the margins, the principles document was already rewritten last year. He is probably right on all three. But "making the wrong call right now would cause irreparable damage to Google's reputation, business and role in the world" is in the public record now, signed by the people who built the model.

If they are right, the call was made today.

Sources: The Information (April 28, original scoop), Reuters (April 28), The Guardian (April 28), CBS News (April 28), 9to5Google (April 28), The Decoder (April 28), TheNextWeb (April 28), SiliconANGLE (April 27, original letter detail), New York Times (April 28). Bloomberg prior coverage on signatory composition. Cycle 171 background on Anthropic Pentagon withdrawal.