Your Local Claude Code Conversations Are Now Part of the Corporate Audit Log — Anthropic Opens a Read Path Into "Local Sessions"
On August 11, Anthropic expanded its Compliance API for auditing so that session records from Claude Code and Cowork running on employees' machines can be read out. Agent conversations that once stayed local now fall within the scope of corporate audits — here's how it works, and the tension between
Interactions with AI agents that used to stay on a developer's machine are now being pulled into the scope of corporate auditing. On August 11, Anthropic expanded its enterprise auditing API, the Compliance API, so that administrators can read out session records from Claude Code and Claude Cowork running on employees' machines. For now, it's a beta available to Claude Enterprise.
The End of "It Runs Locally, So It's Invisible"
The key point is that what's covered are sessions running on employees' own devices. Claude Code runs on each person's PC as a CLI or a desktop app, and until now its conversation history tended to stay local — making it hard for administrators to collect or export it across the organization. With this change, administrators can retrieve, after the fact, a server-side consolidated transcript even for those local sessions.
And it's not just the conversation text that can be retrieved. Along with prompts and responses, the contents of web searches and tool calls made via MCP, and even the contents of skills and artifacts, come back bundled into a single record. Metadata such as the user's identity, organization ID, session ID, and timestamps can be retrieved as well.
Three Entry Points on Offer
Retrieval uses the existing Compliance Access Key and the read:compliance_user_data scope as-is, and is carried out through three endpoints.
| Endpoint | Role |
|---|---|
GET /v1/compliance/apps/sessions/local | List local sessions within the organization (filtering by user is not possible; the whole organization is in scope) |
GET /v1/compliance/apps/sessions/local/{session_id} | Retrieve the metadata for a single session |
GET /v1/compliance/apps/sessions/local/{session_id}/messages | Retrieve that session's conversation transcript |
The design philosophy is that there's no need to build new authentication infrastructure or a separate logging pipeline — it can be handled as an extension of the same mechanism already used to audit Claude chats.
The Business Impact: Falling Under the eDiscovery and DLP Net
This capability matters most for organizations that are required, as a matter of business, to retain records or produce evidence. Anthropic cites electronic discovery (eDiscovery), data loss prevention (DLP) operations, and audit response as use cases. The significance of consolidating visibility — previously scattered across products (regular chat, Cowork, and Claude Code) — into a single interface is far from trivial. In sectors like finance and healthcare, where being able to trace "who sent what to the AI, and when" is a prerequisite, this has been the very issue that determined whether agents could be adopted at all.
Retention periods were spelled out too. Records of local sessions (Cowork and Claude Code on employee devices) are kept for six years by default, or for whatever finite retention period the organization sets. Remote sessions running in Anthropic's cloud are likewise retained for six years.
The Boundaries to Understand Before Adopting
Because it's still in beta, what's out of scope is also clear. The web version of Claude Code, usage via the Claude Platform (API), and cloud-provider editions such as Bedrock, Vertex AI, and Foundry are not included in this coverage. Note that which of these your organization's usage falls under changes what you can actually audit.
The Tension Behind the "Visibility"
For those doing the governing, this is progress — but there are considerations worth flagging on the other side. Because what's covered is development work running on individuals' own devices, the fact remains that the exchanges employees have with AI day to day become storable and readable by the organization for six years. While Anthropic's announcement foregrounds the data-security benefits to the organization, it says nothing in particular about transparency toward the individual or any framework for consent. How to balance auditability against developers' privacy and psychological safety is left to the operational design of the companies that turn the feature on. Strengthening governance and excessive surveillance are two sides of the same coin, and adoption should go hand in hand with clearly communicating a usage policy.
The more the pattern of "running agents autonomously on your own machine" takes hold, the more unavoidable the question becomes of who holds those records, and how far their reach extends. This move signals that the plumbing to give enterprises their answer to that question is starting to fall into place.
References: Compliance API coverage extends to Claude Cowork and Claude Code (Anthropic) / Claude Platform release notes / Retrieve and delete chats, files, and projects (Claude Platform Docs) / Compliance API (Claude Platform Docs)
