Claude for Government Opens to Civilian Agencies

Anthropic is pushing further into the public sector, even as one of its biggest potential government customers keeps it locked out. The company now offers Claude for Government to US federal and state agencies. The platform has been in open beta since July. The target audience is clear: civilian agencies, not the military.

This is not a strategic preference. The Pentagon has banned Claude, and the legal fight over that ban is still not over.

What agencies actually get

Claude for Government runs in a FedRAMP High environment. FedRAMP is the US government's security authorization program for cloud services, and "High" is its strictest level. It is meant for systems that handle the most sensitive unclassified data. For agencies, that certification is often the entry ticket, not a bonus feature.

Beyond the security baseline, the offer looks much like Anthropic's enterprise product:

Same feature set as enterprise customers. Agencies are not getting a stripped-down version of Claude.
Usage-based pricing with fixed caps. Agencies pay for what they use, but spending cannot run past a set ceiling. For public budgets, that predictability matters.
Department-level controls. Administrators can set budgets and decide which models each department can access.
Audit logs and two-person approval. Sensitive actions require sign-off from two people, and activity is logged.
Local chat storage. Conversations stay on agency devices.

None of this is unusual for enterprise AI. But together it reads like a checklist built for procurement officers and compliance teams. Budget caps, audit trails and dual approval are the kinds of controls that let a cautious agency say yes.

Why the Pentagon is off the table

The civilian focus traces back to March. That month, the Pentagon labeled Anthropic a supply chain risk. The trigger was the company's refusal to drop two restrictions in its usage policy: a ban on mass surveillance and a ban on autonomous weapons.

The fallout spread quickly. President Trump ordered all agencies to stop using Claude. Even so, the military still used the model during the Iran war before removing it.

Then the courts got involved. There were two separate designations against Anthropic, and they have gone in different directions:

Late August: A judge in San Francisco struck down one of the designations, ruling it was unlawful retaliation.
Late September: An appeals court in Washington upheld the other.

So Anthropic has won one round and lost another. The Pentagon ban remains in place.

Signs of a thaw, maybe

There is a possible shift in tone. CEO Dario Amodei had dinner with Trump on Sunday. That meeting suggests tensions between the company and the administration may be easing. But the relationship has already swung sharply this year, and a single dinner does not settle a dispute that has run through two courts. Any warming could prove short-lived.

The timing of the launch is worth noting too. Anthropic is courting federal buyers at a moment when Washington is paying close attention to AI labs on several fronts, including an FTC probe targeting OpenAI, Anthropic and others. Government is both a customer and a regulator here, and Anthropic has to manage both roles at once.

The Bigger Picture

This launch shows a company routing around a blocked door rather than waiting for it to open. If defense is closed, civilian agencies are the next best market, and the product is clearly designed to remove their usual objections. The emphasis on caps, audit logs and two-person approval suggests Anthropic understands that public-sector buyers care as much about control as about model quality.

The deeper story is about what happens when a lab's usage policy collides with a government customer. Anthropic held its line on surveillance and weapons, and paid for it with a Pentagon ban. That stance fits a broader pattern in which AI companies and the White House negotiate terms, sometimes through informal deals such as the non-binding AI code of conduct tech CEOs recently signed.

It is worth watching three things next. First, whether civilian agencies actually adopt Claude while Trump's order to drop it stands. Second, whether the split court rulings lead to further appeals. Third, whether the Amodei dinner produces anything concrete. Government revenue could also matter to investors, especially after reports of Anthropic's steep operating losses.