Claude Cowork lands on the FedRAMP High desktop as Anthropic pushes its agent into agencies

Days after going GA on mobile and web for enterprise, Anthropic is putting its file-manipulating agent — and Claude Code — into Claude for Government in public beta.

Anthropic is placing Claude Cowork and Claude Code inside Claude for Government in public beta, extending its file-manipulating agent to the FedRAMP High environment used by U.S. federal agencies. The move lands days after Cowork’s mobile and web rollout, and it collapses the distance between a January research preview and a procurement-ready federal tier faster than most enterprise software ever moves.

Cowork, per Anthropic’s product page, “takes real actions on your files and in connected tools,” with approvals required before deletions or significant actions and desktop as the primary surface. Computer use stays in research preview.

The usage data is the tell. Anthropic’s analysis of 1.2 million anonymized Cowork sessions from May 11–31 found business process and operations work at 33.4%, content creation at 16.4%, and software development trailing at 8.7%. The Decoder pegged non-coding use above 90%. “The use of AI for everyday business work is on the rise,” Anthropic told TechCrunch, which is the polite framing of a more pointed reality: the coding-agent vendors have discovered the rest of the office, and the rest of the office is bigger.

Boris Cherny, head of Claude Code, told Fortune that Cowork was built in roughly a week and a half, largely using Claude Code itself. It launched in January to Max subscribers on the $100 and $200 plans and immediately rattled startups clustered in file organization, document generation, and data extraction.

The governance picture hasn’t caught up. Gravitee’s survey of 445 IT and security leaders found 88% had confirmed or suspected agent-related security incidents in the past year, and only around 14% said their agents were deployed with full security and IT approval. “The deployment velocity is dramatically outpacing governance,” one CRO told InfoWorld. “A background agent reading email, files, and calendars while the user is offline requires very precise scoping.” Anthropic itself, at launch, called agent safety “still an active area of development” and acknowledged Cowork remains vulnerable to prompt injection.

Federal accreditation, then, is arriving before the category has agreed on what accreditation should mean.