Claude for Government: How Agencies Actually Buy It — and Where Contractors Fit In

Federal agencies can acquire Anthropic's Claude three ways: through the GSA OneGov agreement announced in August 2025 (nominal $1 access through August 10, 2026), through existing cloud contracts via Amazon Bedrock and Google Vertex AI marketplaces, or through GSA MAS resellers such as Carahsoft. Claude for Government supports FedRAMP High workloads. The contractor opportunity is deployment services, not license resale.

I spent eighteen years in federal acquisition as a Contracting Specialist and Contracting Officer at GSA, IRS, DoD, and DOI, and I have watched a dozen technology waves hit the government. The pattern is always the same: the product announcement gets the headlines, and the acquisition mechanics decide who actually makes money. Here is how Claude reaches an agency contract file — and where a services contractor fits into that flow.

What is the GSA OneGov agreement with Anthropic?

On August 12, 2025, GSA announced a OneGov agreement making Claude available across the federal government for a nominal fee of $1 through August 10, 2026. The deal covers Claude for Government — Anthropic's offering that supports FedRAMP High workloads — and extends to all three branches of government, not just executive agencies.

OneGov is GSA's model of negotiating directly with the OEM instead of letting every agency cut its own deal. For Claude, the mechanics look like this:

• Who can use it. Eligible federal entities across the executive, legislative, and judicial branches can onboard under the agreement's terms.
• What it costs. A nominal $1 fee for the initial access period — the same pricing posture GSA struck in its other OneGov AI deals. The real pricing question arrives at renewal.
• What it includes. Claude for Government for sensitive unclassified work, purchasable through the government channel GSA designates — Carahsoft has published purchase guidance as the ordering vehicle.
• What it is not. A deployment. The agreement puts the license within reach; it does not configure a single agency workflow, write an ATO package, or train a single analyst.

The GSA announcement is the primary source worth reading before you brief a client or a capture team on this.

How do agencies acquire Claude outside of OneGov?

Through contracts they already hold. Claude runs on Amazon Bedrock and Google Vertex AI, which means any agency with an existing AWS or Google Cloud vehicle can consume Claude as a marketplace service under that contract — no new procurement action required. This is the quiet path most production workloads actually take.

From the CO seat, the marketplace path was always the one that moved fastest, because nobody had to open a new contract file. If the agency's Bedrock environment already has an Authority to Operate (ATO), adding a model endpoint is a change request, not a procurement. That is why contractors who understand cloud contract scope — not just AI — win this work.

What is the compliance posture — FedRAMP and impact levels?

Claude for Government supports FedRAMP High workloads, which covers sensitive unclassified data. For DoD, Claude reaches customers through cloud environments authorized at DoD impact levels, and Anthropic has released Claude Gov models built specifically for U.S. national security customers operating in classified environments.

The layering matters, because "is Claude FedRAMP authorized" is the wrong question. Authorization attaches to the cloud service offering and the agency's own ATO boundary, not to the model in the abstract:

1. The platform layer. Amazon Bedrock and Google Vertex AI carry FedRAMP authorizations; government-dedicated regions such as AWS GovCloud support DoD impact-level workloads. Claude inherits the boundary it runs inside.
2. The agency layer. The agency's Authorizing Official still has to accept the model into the system boundary — data flows, logging, retention, and human-review controls all land in the ATO package.
3. The workload layer. CUI can only touch the deployment pattern that satisfies its safeguarding requirements — NIST SP 800-171 on contractor systems, the agency's FedRAMP baseline on government systems. See FedRAMP.gov for the authorization marketplace itself.

DoD adds the impact-level question — IL4 and IL5 for CUI and national security systems data, IL6 for classified. If a client tells you their DoD customer wants Claude on mission data, the first document to pull is not the AI policy. It is the cloud environment's provisional authorization.

Where is the services opportunity for contractors?

The license is a dollar. The deployment is the contract. Agencies onboarding Claude need integration, security assessment, workflow engineering, prompt and evaluation design, training, and governance support — and nearly all of that buys as professional services under existing vehicles, especially GSA MAS SINs 54151S and 541511 — there is no dedicated AI SIN; AI work rides on the IT services SINs you already know.

Across our 70+ proven GSA contract awards, the firms that monetized a platform wave were never the ones reselling the platform. They were the ones who could stand in front of a program office and say "we have deployed this inside an ATO boundary before." The work breaks down into repeatable offerings:

• Deployment and integration. Standing up Claude inside the agency's Bedrock or Vertex tenancy, connecting it to document stores, and configuring access controls.
• Security and ATO support. Writing the control narratives, data-flow diagrams, and POA&Ms that get the model accepted into the system boundary.
• Use-case engineering. Turning "the Administrator said use AI" into three scoped workflows with measurable baselines — the deliverable COs can actually evaluate.
• Governance and compliance. Building the human-review procedures and audit logging that keep the agency inside its own AI policy and, for vendors, inside GSAR 552.239-7001's disclosure expectations.
• Training and adoption. The line item every agency underfunds and every successful deployment quietly depends on.

How should a services contractor position for this work now?

Get onto the vehicles agencies will use to buy the services — GSA MAS under the IT professional services SINs is the floor — then build one documented, referenceable Claude deployment, even if it is internal. Agencies evaluating AI services in FY26 are asking for demonstrated deployments, not slideware capability statements.

When I sat on the other side of the desk as a GSA Contracting Officer, the technical evaluations that moved fastest were the ones where past performance answered the evaluation criteria directly. For AI services orders, that means:

1. Hold the right SINs. 54151S (IT professional services) carries this work; make sure your awarded labor categories say AI/ML in words an ordering CO can score.
2. Document a deployment. One real implementation — architecture, security controls, measured outcome — beats ten pages of AI vision language.
3. Speak the compliance language. Proposals that reference FedRAMP boundaries, impact levels, and NIST SP 800-171 by name survive technical evaluation; proposals that say "secure AI" do not.
4. Watch the renewal cliff. The $1 OneGov pricing runs through August 10, 2026. Agencies that adopted at scale will need renewal support, alternatives analysis, and cost modeling — a consulting product in itself.

What Is the Bottom Line?

• Three paths, one destination. OneGov for enterprise access, cloud marketplaces for production workloads, MAS resellers for Schedule orders with services attached.
• The compliance question is the boundary, not the model. FedRAMP High via the cloud platform, DoD impact levels for defense work, Claude Gov for classified environments.
• The money is in services. Deployment, ATO support, workflow engineering, governance, and training — sold under SIN 54151S (and 518210C for cloud-based AI).
• Build one referenceable deployment before FY26 buying season. Demonstrated past performance is the discriminator ordering COs are already asking for.
• Mark August 10, 2026. When the nominal pricing period ends, the renewal and alternatives-analysis work begins.

Frequently Asked Questions

Is Claude FedRAMP authorized?

Claude for Government supports FedRAMP High workloads, and Claude is available through FedRAMP-authorized cloud platforms including Amazon Bedrock and Google Vertex AI. Authorization attaches to the cloud service boundary and the agency's own ATO, so the deployment environment — not the model name — answers the compliance question.

What did the GSA OneGov agreement with Anthropic actually do?

Announced August 12, 2025, it made Claude available to eligible federal entities for a nominal $1 fee through August 10, 2026, with Carahsoft publishing the purchase guidance. It pre-negotiates access and terms; it does not deploy anything inside an agency.

Can DoD components use Claude on CUI or classified data?

CUI workloads require a cloud environment authorized at the appropriate DoD impact level (IL4/IL5), such as government-dedicated regions. For classified environments, Anthropic has released Claude Gov models built for U.S. national security customers. The component's Authorizing Official makes the final call within its accreditation boundary.

Can I resell Claude on my own GSA Schedule?

Only with an authorized reseller arrangement from Anthropic or its distribution channel, plus a Letter of Supply supporting the product SIN. For most services firms the better play is delivering deployment and integration services under SIN 54151S rather than chasing thin resale margin.

Which GSA SINs cover Claude deployment services?

SIN 54151S (IT Professional Services) covers integration, security, and AI/ML implementation services; SIN 518210C covers cloud and cloud-related services including cloud-based AI; SIN 541511 covers custom development. There is no dedicated AI SIN. Many agencies also route this work through OASIS+ and agency IDIQs.

Does the GSA AI clause apply to contractors deploying Claude for an agency?

If your GSA contract involves developing or deploying AI, expect GSAR 552.239-7001's disclosure and safeguard requirements to apply, and expect ordering activities to ask about AI use even where the clause is not yet in your contract. Document your AI toolchain now.

What happens when the $1 OneGov pricing ends in August 2026?

Agencies will face renewal negotiations at commercial-scale pricing, which triggers alternatives analysis, usage-data reviews, and budget justifications. Contractors who can model consumption costs and compare deployment options will find a distinct consulting lane in FY27 planning.

If you want your firm positioned on the right Schedule with the right SINs before agencies start awarding this work, our team handles exactly that — start with our GSA Schedule services page.