Claude Compliance API Now Covers Cowork and Claude Code

Anthropic announced August 11 that its Compliance API now covers Claude Cowork and Claude Code, pulling session transcripts from those products into the same audit feed enterprises already use for Claude chats. Coverage is in beta for Claude Enterprise organizations — and for the first time, it captures sessions running on users’ own machines.

Key facts

• The Compliance API gains endpoints for Cowork (desktop, web, mobile) and Claude Code (CLI and desktop), all reachable through the existing Compliance Access Key — no new integration to build.
• Each session comes back as a single record combining session content (prompts, responses, tool call content including web and MCP actions, skills and artifacts) and session metadata (verified user ID and email, organization ID, session and per-message IDs, timestamps).
• The endpoints are additive: existing data pulls are unchanged.
• Coverage does not yet include Claude Code on the web, Claude Code accessed through the Claude Platform, or sessions on Amazon Bedrock, Google Cloud Vertex AI, or Microsoft Foundry.

Why this closes a real gap

For most of 2026, the Compliance API covered claude.ai conversations but not the agentic products that do the actual work on enterprise machines. Third-party security vendors documented the hole repeatedly: Cowork activity was excluded from Audit Logs, Compliance API, and Data Exports, leaving compliance teams unable to produce a record of what an agent touched on a specific workstation.

The August 11 release changes that for on-machine sessions — the surface with shell access, file access, and MCP connectors. Auditors and eDiscovery teams can now answer the question “what did this agent do on this user’s machine at this time” without standing up separate logging infrastructure. Anthropic frames it as closing a gap so Cowork and Claude Code “show up alongside Claude chats.”

The mechanics

The new endpoints return consolidated, server-hosted transcripts per session, so prompts, responses, and tool activity come back together. Verified user identity is included — email addresses and user IDs are confirmed at session level, which is what SIEM ingestion and eDiscovery workflows need for attribution.

Under the hood, the release adds local-session endpoints under /v1/compliance/apps/sessions/local: a list endpoint, a metadata endpoint, and a messages endpoint returning the transcript, all authenticated with the existing Compliance Access Key and the read:compliance_user_data scope. Compliance endpoints share a rate limit of 600 requests per minute per parent organization. The shape of the API matters as much as the coverage: it reuses the access model governance teams already provisioned, so security teams are not maintaining a second credentialing flow.

Anthropic also keeps OpenTelemetry as a complementary channel. Organizations already exporting OTel data can keep it running alongside the Compliance API without extra infrastructure on their side — the API is the audit trail, OTel is the runtime telemetry.

Why the timing makes sense

The rollout arrives just as autonomy becomes the default. On August 7, Anthropic announced that auto mode — where Claude Code acts without per-step approval — becomes the default for Pro, Max, and Team plans on August 14. Sessions that run without human approval are precisely the ones compliance teams need transcripts for. On-machine coverage for audit purposes is the governance counterpart to unsupervised execution: if agents no longer ask permission on every step, the record of what they did becomes the primary control.

Sequence of moves

The pattern is deliberate: on August 3, Anthropic added transcripts for Cowork sessions started on claude.ai web or mobile (cloud-run sessions). On August 11, coverage extended to local sessions on users’ machines. Governance coverage is being rolled out surface by surface, with local agent execution — the riskiest surface — arriving last but now present.

What to watch

• Whether beta coverage extends to Claude Code on the web and the Platform surfaces left out of this release.
• Whether Bedrock, Vertex, and Foundry deployments get the same session-level feed, given enterprises run Claude through those clouds at scale.
• Adoption by SIEM and eDiscovery partners, which determines whether the audit trail is actually used or just available.

Related: Always-on agents compared, Claude Code vs Copilot agents, Daily AI Brief: August 10.