Anthropic has made its Claude for Government product fully available to federal and state agencies, moving it out of a test phase that started in July. Agencies get the same tools as regular paying business customers, but with billing built for how government actually works: no fee per person, just a fixed amount of spending that cannot be exceeded, split up by department so one office cannot blow through another office's budget.
The software runs inside a security setup called FedRAMP High. That is the strictest approval level the US government gives to cloud software, and it is what lets agencies use Claude for sensitive but unclassified work, things like reviewing contracts, drafting memos, or handling casework.
One customer is missing from the list: the Pentagon. The Department of Defense has labeled Anthropic a "supply chain risk," a designation that blocks the military and any defense contractor from using Claude at all. An appeals court upheld that label on September 25, just days before this civilian launch.
The root of the fight is a contract. Anthropic signed a 200 million dollar deal with the Pentagon last year, but talks over how the military could deploy Claude collapsed when the Pentagon asked for broad rights, including use in fully autonomous weapons that pick targets without a human and in domestic surveillance.
Anthropic refused both, arguing its models are not reliable enough for life or death decisions and that mass surveillance of citizens crosses a line it will not help with. Defense Secretary Pete Hegseth accused the company of trying to seize veto power over military decisions. Anthropic sued, won once with a judge calling the label "Orwellian," then lost that win on appeal.
While that fight plays out in court, Anthropic's business with the rest of the government has grown fast. The General Services Administration struck a deal letting agencies buy Claude for one dollar per agency for a year, and agencies including NASA, Treasury, the State Department, and Health and Human Services have already rolled it out for staff. California went further, signing a statewide deal giving all its state agencies access to Claude at half price, the first agreement of its kind for an entire state government.
This is worth paying attention to beyond the drama. Anthropic is showing that an AI company can say no to its single largest potential customer over how the technology gets used, and still keep growing everywhere else. That is a new dynamic for anyone buying AI tools: vendors now have publicly stated limits on what they will build, and those limits can collide with what a powerful customer wants.
If your business touches government contracting in any way, the practical lesson is that vendor choice is no longer just about price and features. It is also about which lines a vendor will or will not cross, because those lines can determine whether you are even allowed to use their product on certain contracts. Companies that work with the Department of Defense specifically now need to check whether their AI tools are built on Anthropic's models at all, since the blacklist extends to contractors.
For everyone else, the bigger pattern is simpler. Government is becoming one of the fastest-growing customers for AI software, and the vendors willing to compete hardest for it, on price and on flexible deployment, are shaping how the rest of the market will eventually buy this technology too.