Google has granted the U.S. Department of Defense access to its AI on classified networks under terms that allow essentially all lawful uses, stepping into a contract Anthropic publicly refused. The deal makes Google the third frontier AI company to take the Pentagon's open-ended terms, following OpenAI and xAI. It also lands while 950 of Google's own employees are circulating an open letter asking the company to do the opposite.
The dispute that opened the door for Google began with Anthropic. The Pentagon wanted unrestricted use of frontier AI; Anthropic wanted contract language ruling out domestic mass surveillance and autonomous weapons. When Anthropic refused those use cases, the DoD branded it a 'supply-chain risk,' a designation normally reserved for foreign adversaries.
That fight is now in court. Anthropic sued, and last month a judge granted an injunction against the supply-chain-risk label while the case proceeds. The injunction does not resolve the underlying contract dispute, but it has temporarily stopped the Pentagon from formally treating a U.S. AI lab the way it treats hostile foreign suppliers.
Key facts
- 01Google granted the U.S. Department of Defense access to its AI on classified networks, allowing essentially all lawful uses.
- 02950 Google employees signed an open letter urging the company to refuse the deal without Anthropic-style guardrails.
- 03Anthropic refused the same terms and was branded a 'supply-chain risk' by the DoD, a label usually reserved for foreign adversaries.
- 04A judge last month granted Anthropic an injunction against the designation while its lawsuit against the DoD proceeds.
- 05OpenAI and xAI have already signed their own DoD deals, making Google the third AI firm to step into the gap Anthropic left.
Google's agreement, according to The Wall Street Journal, contains language saying the company does not intend its AI to be used for domestic mass surveillance or in autonomous weapons. That phrasing is close to what OpenAI agreed to in its own DoD contract. Whether 'does not intend' is legally binding or enforceable, the Journal notes, is unclear.
“Google entered the Pentagon deal despite 950 of its own employees signing an open letter asking it to follow Anthropic's lead and demand the same guardrails.”— Jaeden Schafer
That ambiguity is the gap Anthropic refused to sign into. Anthropic wanted hard prohibitions, not statements of intent, and was willing to walk away from the contract and absorb the supply-chain-risk hit rather than accept softer language. Google, OpenAI and xAI have all decided the softer language is good enough.
The 950-employee letter inside Google asks leadership to follow Anthropic's lead and refuse to sell AI to the Defense Department without comparable guardrails. Google did not respond to a request for comment. The company has not said publicly how it weighed the internal opposition against the contract.
The commercial logic is straightforward. The Pentagon is one of the largest single buyers of advanced computing in the world, and classified-network access is a category where only a handful of vendors can operate. With Anthropic locked out and litigating, the addressable share of that budget effectively expands for whoever signs.
The political logic cuts the other way. Google spent years rebuilding its defense business after the 2018 employee revolt over Project Maven, the Pentagon image-recognition contract the company eventually declined to renew. The new DoD agreement, paired with a fresh open letter from nearly a thousand staff, suggests that internal fight is restarting under different management and a different administration.
It is also worth watching how enforceable the 'no mass surveillance, no autonomous weapons' language turns out to be in practice. The WSJ flagged the question directly: contract intent and contract obligation are not the same thing, and the DoD has shown with Anthropic that it views unrestricted access as the baseline it expects from frontier vendors. If a future use case tests those clauses, Google's contract is the document that will be read.
Anthropic's bet is that being the lab that said no is a long-term asset — with enterprise customers, with regulators, and with the safety-conscious researchers it competes with OpenAI and Google to hire. Google's bet is that the Pentagon revenue and the strategic positioning inside U.S. defense AI are worth absorbing another round of internal dissent. Both bets will be settled by which clauses, in which contracts, get tested first.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
