Claude for Government: Deployment Routes, Admin Set-Up and the GSA Discount

Anthropic says it delivers near-feature parity with its commercial offerings. Here's what the GSA agreement changes, the three ways agencies can deploy Claude, and how the admin side is built, from the tenant down to the individual seat.

What the GSA agreement changes for agencies

The discount is an agreement through GSA's schedule, not a change to list prices, and it covers government partners. Anthropic says that, for the first time, it applies across every product surface, both when agencies buy direct today and through cloud providers in the coming months. It also extends to state, local, tribal and territorial governments.

• Agencies can buy at the OneGov discount through GSA's Multiple Award Schedule today.
• There's no minimum spend commitment. Agencies pay for the usage they consume.
• Anthropic estimates the discounts will save the government $1 billion, as of October 2026.
• Anthropic also plans a dedicated programme to help agencies implement and get more from Claude in the coming weeks.

For context, Anthropic says approximately 500 US federal, state and local government organisations use Claude, and that weekly Claude for Government active users are up about eightfold since February. Those are Anthropic's own figures.

Three ways agencies can deploy Claude

Route	Where it runs	Built for
The Claude API	AWS and Google Cloud, with authorizations up to FedRAMP High and IL6	Agencies building their own tools
Claude Gov models	Classified environments on AWS	National security missions
Claude for Government	A FedRAMP High authorized environment	Federal, state and local agencies

Through Amazon Bedrock, Claude is authorized for FedRAMP High, with select models authorized for DoD IL4 and IL5 in AWS GovCloud. On Google Cloud Vertex AI it's authorized for FedRAMP High. Claude Enterprise is in process for FedRAMP Moderate authorization through the FedRAMP 20x programme, so if you've been comparing it, my guide to Claude Enterprise explains how that plan works for commercial buyers.

The use cases Anthropic lists on its government page include joining datasets across departments to build national programmes, modernising legacy technology such as a COBOL backlog, fixing vulnerabilities in mission systems with a security team approving every patch, and a plain-language assistant for residents that hands eligibility decisions and incomplete cases to a caseworker.

Tenants, organisations and users

Claude for Government is organised in three layers: the tenant, then organisations, then users. There's exactly one tenant per agency. It holds the single sign-on connection to your identity provider, automatic user provisioning through SCIM, verified email domains and the pool of credits and seats.

An organisation is a group of users sharing a pool of seats, policies and spend caps, for example one per bureau, office or programme. Most agencies start with one organisation and add more when they need separate billing accounts, separate usage reporting, or different product settings and access rules.

Every user belongs to exactly one organisation. Their role, User, Owner or Primary Owner, controls what they can manage. Their seat tier, a named bundle of usage limits, controls how much they can use Claude. A user with no seat can sign in but can't send messages until an owner assigns one.

Billing accounts, spend caps and seat tiers

A billing account holds a prepaid credit balance and a pool of seats for each tier. Every organisation links to exactly one billing account and draws from its balance, and several organisations can share one. Tenant admins divide the seat pool and can set a spend cap for each organisation.

A spend cap is a policy limit on how much one organisation can spend from its billing account in a rolling window. Setting one doesn't move any credit, and the account balance is always the ultimate limit. Caps apply to self-managed seat tiers; Anthropic-managed seat tiers are covered by the seat price and don't count against a cap.

Seats go out in two steps. Tenant admins give each organisation a number of seats per tier, then organisation owners assign those seats to people. Changes take effect straight away.

Rolling out Claude Desktop to staff

Claude for Government Desktop is generally available. It gives staff Claude and Claude Code, built on the same application commercial customers use, through a FedRAMP High authorized environment. Members work in the Desktop app, with Chat, Cowork and Code, deployed by IT, and use the browser only to sign in, view their account and administer.

One managed setting connects the standard Desktop app to Claude for Government. Until a device has it, the app shows the regular claude.ai sign-in, which is the first thing I'd check if staff report the wrong login screen. The admin portal sits in the browser, at your Claude for Government address followed by /gateway, not inside Claude Desktop. The first administrator can sign in before single sign-on is connected, through an emailed single-use link.

If you're moving from the older Claude for Government Web, three things change. Conversations and projects are stored on each user's device, so projects aren't shared between people. Each user can run an import themselves. And roles and admin settings aren't copied, so administrators set up the portal again from the setup wizard.

October changelog items admins should know

• The Compliance API now returns only the most recent 30 days of activity, so export anything you need to keep longer.
• Seat tiers gained a Default model setting.
• A Scheduled tasks setting, on by default, can switch off scheduled tasks in Cowork and Code.
• The Code tab asks before each Terminal command, even in bypass-permissions or Auto mode.

Setting up Claude for Government, in order

1. Anthropic sets up a billing account with prepaid credit and a seat pool for the agency.
2. Tenant admins create organisations, connect single sign-on and divide seats and spend caps.
3. Organisation owners assign seats to people.
4. IT deploys Claude Desktop with the managed setting that points it at Claude for Government.
5. Each user signs in with their organisation and, if they're migrating, runs the import from Claude for Government Web.

My verdict on Claude for Government

If your agency was already weighing Claude, the GSA agreement removes two of the usual blockers: the discount and the lack of a minimum spend. I'd start with a single organisation, as most agencies do, and only split into more once you need separate billing or reporting. I'd set spend caps per organisation before seats go out, and I'd tell staff migrating from the Web version that their projects won't be shared any more. For the security review, my guide to Claude Security is worth adding to the reading list.

If your team has the licences and now needs people using Claude well, that's the kind of thing I help with. I run AI training and workshops for teams, and help businesses with AI enablement, AI automation and GEO, which is about getting your business seen and cited in AI answers. Most of that work is with SaaS marketing teams, telecom companies and small businesses. Here's how we can work together.

Frequently asked questions

What is the GSA discount on Claude?

Under the agreement announced on 27 October 2026, government partners get a discount on Claude Enterprise, Claude for Government and the Claude API, bought through GSA's Multiple Award Schedule, with no minimum spend commitment.

Is Claude for Government FedRAMP High authorized?

Yes. Claude for Government is delivered in a FedRAMP High authorized environment and became generally available on 30 September 2026.

Does the GSA agreement cover state and local governments?

Yes. Anthropic says it extends to state, local, tribal and territorial governments.

Are projects shared between users in Claude for Government?

No. In the Desktop version, conversations and projects are stored on each user's device, so projects aren't shared between people.