Claude Code and Claude Cowork arrive in the U.S. government perimeter

Anthropic launches Claude Code and Claude Cowork in public beta within Claude for Government Desktop under a secure, FedRAMP High authorized environment.

Anthropic is launching Claude Code and Claude Cowork in public beta within Claude for Government Desktop, using the same application base as its commercial clients but served from a FedRAMP High authorized environment. One allows public sector teams to build and modernize public service software systems; the other works directly on desktop files, enabling Claude to draft memos, review tenders, process files, or prepare presentations. Inference runs within the FedRAMP High perimeter, and conversation history remains stored locally on the agency-managed device.

The core of the announcement focuses on governance. Budget-wise, an agency can tie AI spending to appropriated funds via standard seats or define its own tiers with spending caps and model limits, with usage purchased in fixed increments with a strict ceiling not to be exceeded; the admin console tracks consumption per user and per model, with alerts before the balance is depleted. Organizationally, a departmental administrator allocates seats and prepaid usage among sub-agencies, each managing its own users, with debit rules, caps, and authorized models defined by SCIM group mappings.

The supervision aspect is highlighted. Every administrative action is recorded in a hash-chained audit log, viewable within the product, and sensitive operations on Anthropic's side require two-person validation. Exports are limited to usage measurement data, to respond to requests related to ATO procedures and Inspector Generals without moving sensitive material. Anthropic also publishes a FedRAMP secure configuration guide for public access and makes available, under NDA via its trust center, its change notification and a penetration test summary of the new desktop client.