Claude for Government reaches general availability with FedRAMP High authorisation
Anthropic's Claude for Government exits public beta on 30 September 2026, giving US agencies FedRAMP High access to Claude, Claude Code, and Microsoft 365 integration.
On 30 September 2026, Anthropic moved Claude for Government out of public beta and into general availability, making Claude’s coding and agentic capabilities accessible to US federal, state, and local agencies inside a FedRAMP High authorised environment. The beta had only opened on 8 July 2026, so this is a brisk turnaround from preview to production-ready.
For agencies that have been watching from the sidelines, waiting for the compliance paperwork to settle before committing, GA removes the main reason to wait.
What agencies are actually getting
The core offering is Claude running inside a dedicated government environment hosted within Palantir’s FedRAMP High boundary, delivered through the Claude Desktop app. That means agencies are building on an assessed, authorised package rather than starting an Authority to Operate process from scratch.
Capability-wise, the pitch is parity with commercial customers: memo drafting, RFP review, casework assistance, and similar knowledge work through Claude Chat, plus Claude Code for software development and system modernisation. Two features are rolling out in early access alongside GA:
Claude Code CLI. The command-line interface for Claude Code is now available in early access through the same FedRAMP High environment, with sign-in and inference both running inside the boundary. Agencies interested in joining the early access programme should contact Anthropic’s public sector team directly.
Claude for Microsoft 365. An integration with Word, Excel, PowerPoint, and Outlook (the latter still in beta) is also available via early access. This puts Claude tooling inside the software most agency staff already use every day, administered through the same controls as the rest of the platform.
The billing model is worth understanding
Claude for Government drops per-seat fees entirely. Agencies purchase prepaid usage in fixed increments with a hard not-to-exceed cap, which means spend cannot drift beyond what has been obligated. There is no seat minimum.
For procurement officers, that structure maps reasonably cleanly onto how government contracts work: you know the ceiling before you sign, and you are not committed to a headcount figure that may not reflect actual usage patterns.
Administrators get granular control over how that prepaid budget is allocated. Departments can be set up as sub-agencies, each with their own administrators, user tiers, model limits, and dollar caps. Burndown alerts flag when a balance is running low, and usage exports give metering data by user and by model.
The governance controls that authorising officials will care about
Audit logging is tamper-evident and hash-chained, reviewable directly in the admin portal alongside metering-only usage exports. Agencies connect their own identity provider for single sign-on, with SCIM group mappings handling rate limits, spend caps, and model permissions per tier. The FedRAMP Secure Configuration Guide is available through Anthropic’s trust centre, and the application deploys through standard agency MDM platforms.
One detail worth flagging for records and legal teams: conversation history is stored locally on the agency-managed device, not in Anthropic’s infrastructure. That keeps data within the agency’s boundary, but it also means that capturing records, responding to FOIA requests, and honouring legal holds is the agency’s responsibility to manage. If your records management workflows assume server-side storage, this is a point to address before rolling out broadly.
What this means for your agency
If you are in federal or state government and your teams are already using commercial AI tools under informal or shadow arrangements, GA gives you a path to consolidate onto a platform that has the compliance posture your authorising official can actually approve.
The hard-cap billing model removes the open-ended cost exposure that makes CFOs nervous about usage-based AI services. The FedRAMP High authorisation means you are not starting an ATO from scratch. And the fact that no separate cloud-provider relationship is required lowers the procurement friction compared with deploying Claude through a cloud marketplace.
For existing customers who were already on the beta, migration to the Claude for Government Desktop offering is required by 4 October 2026. Conversation history can be imported from the old environment through an in-app import tool.
New agencies can request access at claude.com/solutions/government. For the Claude Code CLI or Claude for Microsoft 365 early access programmes, you will need to go through Anthropic’s public sector team rather than self-serve.
A bit of background
The path to this GA has been methodical. Anthropic introduced dedicated Claude Gov models for classified national security use cases in June 2025. In August 2025, those models landed on the GSA schedule at $1 per agency for the first year, opening access across all three branches of the federal government. The public beta followed in July 2026, adding Claude Code and the desktop environment. GA, announced on 30 September 2026, is the natural next step rather than a sudden pivot.
Anthropic is also, separately, pursuing a legal challenge to a Department of War supply chain risk designation. A federal court in the Northern District of California ruled on 27 August 2026 that one such designation was unlawful. That case is unlikely to affect day-to-day agency use of Claude for Government, but it is context worth knowing if your agency’s legal team asks about it.