Anthropic made Claude for Government generally available to federal and state agencies on 30 September 2026, moving beyond a July public beta with no seat fees and a hard cap on usage spending, Unite.AI reported.
Key points
- Agencies buy usage in fixed increments, capped at the amount they have obligated.
- Claude for Government runs through a FedRAMP High authorised environment, with conversation history stored on agency-managed devices.
- Claude Code’s command-line interface and Claude for Microsoft 365 are entering early access.
Fixed increments replace Claude seat fees
Under the general-availability terms, agencies purchase usage in fixed increments rather than paying for each seat. A hard not-to-exceed cap keeps spending within the amount an agency has obligated. Procurement officers can contract directly with Anthropic, which remains the contracted and billing party.
Administrators can divide prepaid usage among departments. Department-level administrators can then assign it to sub-agencies, which manage their own users. Administrators can also set user tiers with spending and model limits, monitor consumption by user and model, and receive alerts as a balance runs low.
The public beta, launched on 7 July 2026, offered standard seats as well as custom tiers with separate spending and model limits. Its billing options allowed programme offices to align AI expenditure with appropriated funds. General availability replaces seat fees with purchases of usage under the spending cap.
Agency identity systems provide single sign-on, with setup available through an administrative portal. SCIM group mappings assign rate limits, dollar caps and permitted models to seat tiers. Administrators can also set defaults for sub-agencies, including which connections and features they may use.
Agency devices retain Claude conversation history
Claude for Government is delivered through a FedRAMP High authorised environment. Anthropic said agencies would receive capabilities comparable to those available to commercial customers while meeting compliance requirements, with new features generally arriving on the commercial release schedule.
Conversation history remains on agency-managed devices. Audit logs capture administrative actions for review by organisation administrators, while usage exports contain metering data only. Security teams and authorising officials receive logs and documentation for each agency’s authority to operate process.
On Anthropic’s side, sensitive operations require approval by two people. The July beta included tamper-evident, hash-chained audit logs and configuration defaults that could carry instructions governing Claude’s interactions with users. Inference in that beta ran inside the FedRAMP High authorised environment.
Anthropic made its FedRAMP Secure Configuration Guide public at the beta launch. It also provided a formal change notification and, through its trust centre under a nondisclosure agreement, a summary of penetration testing for the desktop client.
Claude Code and Microsoft 365 enter early access
Anthropic is bringing Claude Code’s command-line interface and Claude for Microsoft 365 into early access in the government platform’s existing environment, subject to its administrative controls. The wider government platform gives staff desktop access to files and to skills, plugins and projects for tasks including memo preparation, RFP reviews and casework.
Anthropic released Claude Code and Claude Cowork in public beta for Claude for Government Desktop on 7 July. At that stage, Claude Cowork could work with agency files on the desktop for tasks including memos, RFP reviews, casework and presentations.
Agencies can deploy the desktop application through standard mobile device management platforms without a separate cloud-provider relationship. Existing customers can move their conversation history into the application using an in-app import.
