The Pentagon's chief AI officer just confirmed on the record what we reported on Tuesday. Google now has classified Pentagon access to deploy its AI "for any lawful governmental purpose," while Anthropic remains formally excluded from Defense Department contracts and is fighting it in court. Three of the four largest American AI labs now run on military networks. The fourth is the one that asked for safety guarantees and got blacklisted.

What changed in the last 24 hours

On Monday afternoon, 600+ Google and DeepMind employees signed an open letter to Sundar Pichai and Google leadership asking the company not to sign the classified Pentagon expansion. Within 24 hours, Pichai's office signed it anyway. We covered both moves in real time. Tuesday afternoon, the Pentagon's AI chief publicly confirmed the deal in interviews with CNBC and the New York Times. Wednesday, Google's letter to Defense lawmakers and the Pentagon's own statement made the language explicit: any lawful governmental purpose. No carve-outs. No monitoring exceptions of the kind Anthropic asked for and was punished for asking.

The contract is part of an existing $200 million Google-Pentagon agreement first signed last year. The expansion to classified networks is the new piece. Gemini was already on Pentagon unclassified systems serving roughly 3 million Defense Department personnel. Now it can run on classified ones. Same model, different security perimeter. That is the whole story in one sentence.

The scoreboard

OpenAI: signed the day Anthropic was dropped. Three voluntary red lines (no autonomous lethal targeting, no foreign-leader influence ops, no chemical/biological/nuclear weapons design). xAI: signed within weeks. Public via Musk's X account. Google: signed Tuesday. No public red lines. Just "any lawful governmental purpose." Anthropic: dropped February for asking for safety carve-outs. Now formally excluded from DOD contracts pending litigation. Allowed to keep working with civilian government agencies during the case.

Three labs in. One out. The lab that is out is the one whose entire brand promise is that safety constraints come before commercial deployment. The lab that just won the biggest classified expansion is the one whose own employees publicly asked it not to sign.

Why "any lawful governmental purpose" is the line

The Information's original scoop on the contract language is the document everyone has been quoting. The phrase "any lawful governmental purpose" is what the Google open letter specifically objected to. It is unrestricted. It does not exclude lethal targeting. It does not exclude foreign influence operations. It does not exclude weapons-design assistance. The standard is whatever a federal lawyer signs off on. That is why 600 employees said it would cause irreparable damage to Google's reputation. That is why Pichai signed it anyway.

Compare to OpenAI's three voluntary carve-outs. Compare to Anthropic's request for similar carve-outs that triggered the blacklisting. Google's terms are functionally identical to having no terms. Three of the four major frontier AI labs in America now operate inside the Pentagon's classified networks under contracts that did not require the safety language Anthropic refused to sign without.

The talent risk is now real

Two ex-DeepMind founders raised over $2 billion in the last six weeks specifically positioning their startups as alternatives to Big Lab military entanglement. David Silver's Ineffable Intelligence pulled $1.1 billion. Koray Kavukcuoglu's AMI Labs pulled close to $1 billion. Both pitches included the words "safety alignment first." Both came after the Anthropic blacklisting and before the Google deal. The recruiting brochure is now written for them by their former employer.

Watch the next 30 days. If the Google open letter signatories include senior research staff, expect resignations. If those resignations land at Ineffable, AMI Labs, or Recursive Superintelligence (the new Demis Hassabis-orbit lab raising $1 billion per CNBC), the safety-first lab category gets the strongest hiring window of 2026.

The Anthropic IPO complication

Anthropic is preparing an October IPO at a reported $800 billion+ valuation. Its S-1 must disclose the Pentagon blacklisting as a material risk factor. It also has to disclose that the blacklisting was triggered by the company's own safety stance. That dual disclosure is unusual: the company is essentially telling investors "we lost a contract because we refused to abandon our safety brand, and we are now pricing this IPO around that brand." That math works only if enterprise buyers reward the safety brand harder than military buyers punish it. Wednesday's confirmation that Google won the classified expansion makes the test sharper, not gentler.

What this resets

From here on, when somebody says "American AI labs are aligned around safety," check the Pentagon contract list first. Three of the four most consequential labs in this country are now classified-network defense contractors. The one that isn't is in court trying to get back in. The era where you could be the moral authority on AI safety AND the prime contractor on classified military AI is closed. Pichai picked which side of the line Google sits on. The 600 employees who tried to stop him just lost the most public internal vote in modern Big Tech history.

Sources: CNBC Pentagon AI chief interview (April 29), TechCrunch original Pentagon expansion coverage, New York Times "Google Signs A.I. Deal With the Pentagon" (April 28), NBC News Pentagon AI services confirmation, our own Cycle 172 and Cycle 171 Google Pentagon coverage, The Information original scoop on "any lawful governmental purpose" contract language.