Build2 publishers3 min readPublished
Anthropic sells FedRAMP High Claude to US agencies in prepaid blocks under a hard cap
Anthropic made Claude for Government generally available to US federal and state agencies under FedRAMP High, sold in prepaid blocks with no seat fees. A hard cap fixes the bill, so teams now plan for blocks running out mid-month and for chat records kept on agency devices.
The Engineer · Build desk

What happened
- Administrators set budgets and model access per department, sensitive operations need two-person approval, and audit logs are kept.
- Conversation history stays on agency-managed devices, and ExecutiveBiz reports that compliance exports carry only usage-metering data.
- The Claude Code command-line tool and Claude for Microsoft 365 are entering early access in the same FedRAMP High environment.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- constraint A team that exhausts its prepaid block stops working until more is bought, so per-team and per-model caps and usage monitoring become part of the rollout plan.
- decision Agencies that want the GSA OneGov vehicle have 31 days from launch, so the contract route has to be chosen within weeks.
- contradiction Tech AI Wire describes the August ruling as voiding the Pentagon blacklisting; The Decoder says it covered one of two designations and an appeals court upheld the other, so defense use stays closed.
"Agencies pay no seat fees for Claude for Government. Usage is bought in fixed increments under a hard not-to-exceed cap," the announcement reads, as quoted by Unite.AI [5]. Seat pricing ties cost to headcount. This model ties it to consumption, with a ceiling the bill cannot pass. The failure mode changes from an overrun to a stoppage. Tech AI Wire's write-up of the launch warns that heavy users can run out mid-month, with caps set by dollar and by model [16].
I think the cap is the right trade for a buyer that has to fix its spend before the work starts. It is also one of the few cloud billing terms a contracting officer and an on-call engineer will read the same way. Administrators set budgets and model access per department, The Decoder reports [7]. Users sign in through the agency's own identity provider, and SCIM group mappings sync groups from the directory [6]. The sources do not say whether a department budget binds to a SCIM group. If it does, provisioning a user also puts that user under a cap. If it does not, the budget hierarchy has to be maintained separately from the directory.
Conversation history stays on agency-managed devices, according to Anthropic [8]. ExecutiveBiz reports that compliance exports carry usage-metering data only, not the content of those conversations [9]. The central plane records who used how much. The content sits on the endpoints, so records retention and backup of past work happen there too, and Tech AI Wire advises asking records officers early how they want local history handled [20]. Sensitive operations need approval from two people, and administrators get audit logs [7].
Anthropic's post dates the release to September 30, 2026, ending a public beta that began in July, while ExecutiveBiz reported the announcement on October 1 [4][1]. The Decoder reports agencies get the same features as enterprise customers [21]. Unite.AI's list covers desktop file handling, skills, plugins, projects, Claude Code and Claude Cowork, plus admin controls [10]. The Claude Code command-line tool and Claude for Microsoft 365 are only now entering early access in the same environment [11]. For now, the version of the coding tool a developer scripts from a terminal sits outside the generally available scope [11]. FedScoop reported in September that Fable 5.1 had been added to the service [12].
Teresa Carlson, Anthropic's global head of public sector, told FedScoop that any agency requiring FedRAMP High can use the service [13]. FedScoop reports she favors a "crawl, walk, run" approach [14]. Agencies that want to buy through GSA OneGov have until the end of October, ExecutiveBiz reports [15]. From the September 30 launch, the window is 31 days [1].
The buyers are civilian agencies [17]. The Decoder reports that the Pentagon bans Claude, having labeled Anthropic a supply chain risk in March after the company refused to drop its bans on mass surveillance and autonomous weapons [17]. A San Francisco judge struck down one of two designations as unlawful retaliation in late August, and a Washington appeals court upheld the other in late September, according to The Decoder [18]. Tech AI Wire's summary mentions only the August ruling [19].
What to watch
- Whether agencies close GSA OneGov purchases before the October 31 deadline, and on what terms.
- When the Claude Code command-line tool and Claude for Microsoft 365 leave early access under FedRAMP High.
- Any further appeal of the Pentagon supply chain designation the Washington appeals court upheld in late September.