The California–Anthropic Claude deal makes the AI assistant the first such tool available to every state agency, city, and county in California, at half the standard price, putting Governor Gavin Newsom’s administration squarely at odds with a federal government that has formally declared Anthropic a security threat.
Chris Given, California’s chief information officer and Department of Technology director, told Politico that the Pentagon’s ‘supply-chain risk’ designation of Anthropic ‘just didn’t come up’ during negotiations. That designation, issued by Defence Secretary Pete Hegseth, is the first ever applied to an American company, according to Lawfare.
Claude Comes to City Hall at Half Price
Under the agreement, state agencies as well as California cities and counties that choose to participate can access Claude at a 50% discount on standard pricing. Anthropic offers a version of Claude specifically built for government clients that provides greater security than its consumer-facing products, according to the Los Angeles Times.
California was not starting from scratch. The state had already been using Claude before the formal announcement, including to build tools encouraging public engagement in AI policy discussions and to assist state workers. The new deal follows a March executive order by Newsom that directed the state to accelerate its use of AI to make government more efficient while maintaining stronger safety standards.
The Pentagon Fight That Newsom’s Team Says Didn’t Come Up
The federal backstory is considerably thornier. The dispute between Anthropic and the Pentagon had been building since January, when Hegseth’s AI strategy memorandum directed that all Department of Defence AI contracts adopt standard ‘any lawful use’ language, according to Lawfare. Anthropic refused to accept that language, seeking explicit protections against using Claude for domestic surveillance of Americans or deploying autonomous weapons without human oversight.
Hegseth declined those conditions. On 27 February, his office formally issued the ‘supply-chain risk’ designation. Anthropic received written confirmation on 4 March, according to Anthropic’s own statement, which also clarified the scope: the designation applies only to customers’ use of Claude ‘as a direct part of’ their contracts with the Department of Defence, not to every government or commercial customer who happens also to have Pentagon work.
That distinction matters, but the practical effect was swift. CNBC reported that defence vendors and contractors were required to certify they do not use Anthropic’s models in their Pentagon work, with the designation communicated to Anthropic’s leadership as effective immediately. Lawfare also noted that the Treasury Department announced it was cancelling its use of Anthropic’s products following the designation.
The Pentagon moved on quickly. On the same day Hegseth’s office issued the designation, OpenAI chief executive Sam Altman announced a deal giving the Defence Department access to OpenAI models on its classified network. Altman stated that OpenAI ‘will build technical safeguards to ensure our models behave as they should’ and that the government agreed it ‘would not force OpenAI to make [the model] do’ a task it refuses.
OpenAI subsequently amended that agreement to include language stating that, ‘consistent with applicable laws,’ its models ‘shall not be intentionally used for domestic surveillance of U.S. persons and nationals,’ and the Pentagon affirmed its services would not be used by department intelligence agencies such as the National Security Agency, according to The Hill. In other words, OpenAI eventually secured protections not far removed from those Anthropic had originally demanded, after Anthropic had already been shut out.
Lawfare’s analysis questioned whether the designation itself will hold up legally. The organisation argued that the Pentagon used the hallmarks of a specific Defence Department procurement statute but that the statute does not reach agencies beyond the department, and that Anthropic received no notice and no opportunity to respond before the directive took effect.
California has charted its own course through all of this. The California–Anthropic Claude deal represents the clearest state-level counterweight to federal AI procurement politics so far, and whether other states follow Newsom’s approach, or whether Anthropic’s legal challenge to the Pentagon designation reshapes the landscape first, will determine just how long this particular divergence holds.
