Claude went to work for government, with the audit trail public services need.

What changed

Anthropic opened Claude for Government Desktop in public beta. Claude Code targets modernising public-sector software and Cowork handles document-heavy work like memos and RFP reviews. Conversation history stays on agency-managed devices, inference runs in a FedRAMP High authorised environment, audit logs are tamper-evident and hash-chained, sensitive Anthropic-side operations need two-person approval, and agencies contract and pay Anthropic directly with optional spending caps.

Why it matters

Government work is where 'who approved this and can we prove it' is not a nice-to-have, and this deployment is built around evidence: local history, metering-only exports, published secure-configuration guidance. It is also a preview of the accountability bar every regulated industry will eventually be held to.

Who should care

• Public-sector teams evaluating AI for casework and code
• Compliance leads in regulated industries watching the accountability bar

What to do

If you are an agency, start with one document-heavy workflow where a named officer already reviews the output, and test whether the audit trail answers your inspector's questions. Everyone else: note which controls government demanded, and ask which of them your own AI use is missing.