TL;DR
- Agency Access: Anthropic’s Claude for Government desktop AI app is generally available to federal and state agencies after its July beta.
- Coding Preview: Claude Code’s early-access command-line interface lets developers type requests for coding work into a text-based window.
- Office Preview: Claude for Microsoft 365, an integration for drafting and editing inside Office apps, is rolling out in early access.
- Spending Controls: Agencies buy prepaid usage without seat fees and can limit spending by department, user group and model.
- Defense Restrictions: A September court ruling keeps restrictions on Claude’s use in defense systems and contract work in place.
AI developer Anthropic moved Claude for Government out of beta on September 30, making its desktop service generally available to eligible US federal and state agencies. It lets staff work with files and develop software in what Anthropic describes as a government-authorized cloud environment.
The service combines everyday office work with coding: agency staff can delegate memo preparation, reviews of requests for proposals and casework involving desktop files, while developers use Claude Code to build or modernize software. Its spending and access controls let departments allocate the service across their bureaus and programs.
File Work and Coding on the Desktop
The July 7 government beta already paired Claude Code with Cowork, Anthropic’s file-work assistant, and introduced departmental spending controls.
Claude Code’s command-line interface is rolling out in early access for the government service. Developers can describe coding tasks in a text-based terminal, and the agentic coding assistant helps with writing code and running tests.
Claude for Microsoft 365, Anthropic’s integration for drafting and editing inside Office apps, is also rolling out in early access through the same government environment and administrative controls. Agencies join either early-access program by contacting Anthropic’s public-sector team. The government app so far has no browser-based experience or mobile access.
How the Available Workflows Compare
Google’s Gemini for Government offers an overlapping route for document work. Its September 24 deployment guidance lists document uploads and Deep Research, its research assistant, as authorized for FedRAMP High deployments under the federal cloud-security program. Agent Designer, a tool for creating AI assistants without writing code, is also listed as authorized. Agencies deploy these resources in an Assured Workloads folder, which groups cloud resources under the required government compliance controls.
Claude’s offering centers on managed desktop file work and included coding. Google’s coding assistant, Gemini Code Assist, is available but is not authorized for FedRAMP High.
Agencies building Claude into their own software have separate managed API routes, which let applications call the model directly. Anthropic lists Amazon Bedrock in AWS GovCloud and Google Vertex AI with Assured Workloads for that purpose. Those cloud services serve application integration, while the desktop app gives employees a ready-made assistant.
Agency Access and Security Approval
New organizations request access through Anthropic’s sales team. Its Claude for Government product FAQ also includes US local agencies and qualifying organizations supporting government work, such as contractors. Agencies can contract directly with Anthropic or through government technology distributor Carahsoft, without needing a separate cloud-provider relationship for the desktop service.
Agency IT teams deploy the app through their managed-device software systems and connect their existing sign-in provider. An employee needs an assigned seat to send messages.
Anthropic’s desktop service is running inside Palantir’s FedRAMP High authorized government environment. Its data scope includes controlled unclassified information (CUI). CUI is unclassified information requiring protection or sharing controls under law, regulations or government-wide policy. Agency officials receive documentation supporting their own authorization-to-operate process, or ATO, for a deployment. Classified Claude workloads use separate deployments through Amazon Bedrock, AWS’s managed AI service.
Conversation history stays on the agency-managed device, while model processing runs in the government cloud environment. Administrative audit logs record configuration and account actions, and usage exports contain metering data rather than conversation content.
Prepaid Usage and Procurement Costs
Agencies can buy usage in fixed prepaid increments, with no per-seat fee or seat minimum. A hard not-to-exceed cap keeps spending within the amount the agency has obligated. Department administrators allocate the balance to sub-agencies, whose user tiers can limit spending, request rates and allowed models.
SCIM, the account-synchronization standard, maps agency directory groups to those tiers. Administrators can track usage by employee and model, and receive alerts as the prepaid balance runs down. Agencies obtain prices from Anthropic’s sales team.
California’s agreement with Anthropic offers Claude through the state’s shared technology-services channel for state agencies and cities and counties that opt in, alongside training and technical assistance.
Jessica Tillipman, George Washington University’s government procurement-law associate dean, warned in an August webinar that accumulated projects and working habits can make switching painful. Users may stay with a tool because migrating their work is disruptive, even when another product is technically available.
Laura Stanton, then a senior purchasing official at the General Services Administration (GSA), argued at the same webinar that pooled agreements can strengthen bargaining power and reduce duplicated purchases and inconsistent terms. She also said agencies would continue to need implementation, integration and cybersecurity partners.
Defense Work Remains Restricted
On September 25, the US Court of Appeals for the D.C. Circuit upheld the Department of War’s exclusion of Claude from its supply chain. The department, previously known as the Department of Defense, had ordered removal of Anthropic products from its systems within 180 days of its March 6 memorandum and barred contractors from using them in work for the department.
The dispute followed Anthropic’s refusal to relax contractual prohibitions on lethal autonomous warfare and domestic surveillance. The court’s majority accepted the department’s concern that restrictions on Claude could interfere with military operations; Anthropic raised the risk of unconstrained models identifying inappropriate targets for lethal force.
For broader federal purchasing, GSA reopened a buying channel for Anthropic in April through its Multiple Award Schedule, which agencies use to purchase technology under government contracts. It continued allowing agency systems to connect to Anthropic products after a March 26 injunction paused implementation of President Trump’s order for federal agencies to stop using the company’s technology.