Claude Code and Cowork Agents Reach FedRAMP High Government PCs

On July 7, 2026, Anthropic launched Claude Code and Claude Cowork in public beta for Claude for Government Desktop. Delivered within a FedRAMP High authorized environment, the release gives U.S. public sector developers and administrators access to specialized agentic tools that run strictly within agency-managed boundaries.

Task Routing with Dispatch

The platform introduces Dispatch, a persistence feature that routes tasks between specialized tools based on intent. Development tasks like refactoring or incident investigation are sent to the terminal-based Claude Code. Administrative tasks, such as generating slide decks or reviewing Request for Proposal (RFP) files directly on the local machine, are routed to Claude Cowork to help users automate desktop workflows.

The Dispatch feature specifically reduces the idle time associated with compiling code or scanning large infrastructure repositories. When a developer initiates a system migration analysis via Claude Code, the agent processes the request independently. If the agent encounters an ambiguous file path or requires credential authorization, Dispatch halts execution and sends a targeted prompt to the user’s mobile device or web interface. This asynchronous model allows heavy workloads to run unattended while maintaining human oversight for critical decisions. For commercial users, Anthropic simultaneously expanded Claude Cowork to web and mobile for “Max” tier subscribers to enable similar cross-device continuity.

Compliance and Architecture

Operating within a FedRAMP High boundary dictates strict data sovereignty parameters. Conversation history is stored locally on the agency-managed device, while inference runs entirely inside the authorized government cloud environment.

Administrators manage deployment through native governance tools designed for federal oversight. Security teams receive tamper-evident audit logs to support the Agency Authority to Operate (ATO) process. IT departments can use SCIM group mappings to allocate seats, configure rate limits, and restrict model access across sub-agencies to securely manage Claude API workloads with identity federation.

This launch represents a structural shift for Anthropic following a volatile first half of 2026. Federal agencies had paused deployments earlier in the year after the company refused to alter contractual bans on autonomous weapons and mass domestic surveillance. The standoff resulted in a brief Department of Defense “supply chain risk” designation, which a federal judge enjoined in March. The new desktop platform is built specifically to re-engage this market through verifiable compliance controls.

Model Configuration and Pricing

The desktop environment supports Claude Sonnet 5 and Claude Fable 5, the latter having cleared temporary export controls in mid-June. Procurement is designed around “appropriation-friendly” billing cycles, allowing program offices to purchase fixed increments with a hard not-to-exceed cap.

Model	Target Workload	Billing Structure	Input/Output per 1M Tokens
Claude Sonnet 5	General reasoning	Seat-based tier limits	Included in seat allocation
Claude Fable 5	Complex agentic tasks	Usage credits	~$10 / ~$50

As of July 7, high-end usage for Fable 5 shifted from weekly quota limits to a pure usage-credit model for government buyers. Program offices can define seat tiers with specific spend caps to prevent budget overruns during long-running autonomous tasks.

If you manage software modernization for federal systems, the availability of terminal-native refactoring tools in a FedRAMP High boundary removes a primary compliance blocker for AI agent adoption. You can now script automation against local files and internal codebase repositories without triggering data egress violations.