Claude Code Comes to Government With Audit Logs Built In

Anthropic put Claude Code and Claude Cowork into public beta inside Claude for Government Desktop this week, running inside a FedRAMP High authorized environment with governance features the commercial product doesn’t ship: department-level spending controls, tamper-evident audit logs built for the Authority to Operate process, and conversation history that stays local to the agency-managed device. It’s the compliance scaffolding a government buyer needs before an agentic tool touches a single real workflow.

The scaffolding matters because the underlying capability is already proven at government scale. Alberta’s Ministry of Technology and Innovation used Claude Code — the same product now shipping with FedRAMP governance — to scan 466 million lines of code across 1,280 applications and 3,400 repositories in about 20 hours, work the ministry estimated would take 6.5 years by traditional means. Roughly 50 agents ran in parallel, and the team rebuilt a subsidy-program portal originally hand-coded in Java 25 years ago — a five-month build — in four to five days, with every patch reviewed and approved by the ministry’s own engineers before it shipped.

What connects the two stories is sequencing, not novelty. Alberta got the capability first and proved it on a five-month problem with willing internal engineers signing off on every patch. The FedRAMP release is Anthropic building the audit trail and spending controls a more risk-averse agency needs before it lets an agent anywhere near procurement or casework. For companies watching how AI adoption actually moves through regulated environments, that’s the real lesson: the technology rarely waits for governance to catch up, but the governance is what turns one ministry’s pilot into a product other agencies can buy.