Our company banned employees from using Cowork for client data a few months ago, citing the lack of an audit trail. Should we lift that ban immediately?
Lifting it immediately isn't recommended — a more solid sequence is confirming three things before adjusting policy. First, confirm whether the Compliance API is even enabled for your organization; if you've never used it, the new session endpoints need to be enabled through your existing Compliance Access Key, and this doesn't happen automatically. Second, confirm whether the deployment actually handling client data falls within this coverage — if it's deployed through Bedrock or Vertex AI, this update doesn't reach it yet. Third, check with your compliance or audit team on whether the current lack of a deletion endpoint for local sessions conflicts with your company's existing data retention or deletion policy.
Once those three are confirmed, lifting the ban is technically sound, since the audit gap has genuinely been closed. But policy changes usually involve internal governance processes too, and it's worth running a formal reassessment rather than changing an existing restriction just because of what one article said.
How does the transcript from the new session endpoints actually differ, in practical use, from the event data already streamed through OpenTelemetry?
The core difference is the shape and stability of the data. What OpenTelemetry streams out is real-time, event-by-event operational data, designed for monitoring dashboards and live alerting, with no guarantee of long-term retention or structural stability. What the Compliance API returns is a complete, server-hosted session transcript, designed for scenarios like audits and eDiscovery where you need to go back and verify exactly what happened at a specific point in time, with a comparatively stable data format and retention mechanism.
In practice, the two are complementary rather than substitutes: OpenTelemetry is suited to asking "is the system operating normally right now, is anything anomalous happening"; the Compliance API is suited to asking "what exactly happened in a specific employee's session three months ago." If your organization needs both, keeping both mechanisms running in parallel covers your compliance needs more completely than relying on just one.
With no deletion endpoint for local sessions yet, what's the actual risk? Does that mean an employee's data can never be deleted?
It doesn't mean the data can never be deleted at all — it means deletion currently can't be executed centrally by an admin through the Compliance API channel. The local conversation history on a user's own computer can theoretically still be handled manually by the user on their own device. What's actually affected is the ability for an enterprise compliance team or admin to centrally, programmatically, and in bulk clear local session records for a specific user or time range through the API — that specific capability is what Anthropic explicitly flags as not yet available.
For organizations subject to data deletion requirements (handling user deletion requests, or operating under industries with explicit data retention limits), this gap is worth taking seriously. It's worth confirming with your compliance team whether this limitation creates an actual compliance shortfall before formally folding local session audit and retention into standard operating procedure, rather than assuming that "we can read it now" means the entire data lifecycle is already under control.
This update was originally labeled a beta for Enterprise customers, then updated to GA — does that gap mean the feature wasn't stable enough yet?
The more reasonable read of that gap is a common product release cadence, not necessarily instability in the feature itself. Looking at the update history on the announcement itself, Cowork and Claude Code coverage was already the core content of the August 11 release; the August 26 update moved that specific coverage from beta to formally GA, while also adding beta coverage for the Microsoft 365 add-ins (Excel, Word, PowerPoint, Outlook) and Claude Science. That means the core Cowork/Claude Code coverage went from beta to GA within about two weeks — a normal maturity-driven upgrade path, not a sign that the core feature broke and got patched during that window.
If your organization treats beta and GA differently in terms of internal risk tolerance (say, beta-stage features aren't approved for regulated production workflows), you can now cite the GA status directly to establish that this feature has met whatever stability threshold your organization typically requires, without needing to wait for a further status change.
If you've looked into Claude Cowork's compliance posture recently, you've likely run into a claim that kept repeating: Cowork sits entirely outside Anthropic's three compliance mechanisms — Audit Logs, Compliance API, and Data Exports — meaning a security team couldn't pull a report showing what files a given user's Cowork session touched. That claim was accurate through the first half of 2026, but it's now out of date. Anthropic announced on August 11 that Compliance API coverage now extends to Cowork (desktop, web, and mobile) and Claude Code, and updated the post on August 26 to confirm this coverage is now generally available, no longer in beta. This piece breaks down what actually changed, what it means in practice for Enterprise teams, and which gaps are still open.
Before this update, Cowork's conversation history lived only on a user's own local machine, wasn't subject to Anthropic's standard data retention policies, and couldn't be centrally managed or exported by admins. That meant if an employee used Cowork to handle sensitive data, there was no centralized record a company could go back and audit. The only visibility available came from streaming Cowork events to your own SIEM via OpenTelemetry — but that channel is fundamentally operational monitoring data, not a stable audit-record format, and by default it ships user prompt content in plaintext, meaning setting up that monitoring pipeline also meant shipping sensitive content unaltered to another system.
Per Anthropic's official blog post, the new session endpoints return a consolidated, server-hosted transcript for each Cowork or Claude Code session, bringing prompts, responses, and tool call activity together into a single session record. Each record carries two kinds of data: session content (prompts and responses, tool call content covering both web and MCP, and skills and artifact content, all captured as transcript text) and session metadata (verified user ID and email address, organization ID, session and per-message IDs, and timestamps). That means enterprise compliance teams can now actually pull a record at the level of "what tools did Claude call, and what content did it access, in this specific user's Cowork session at this specific time."
There's a distinction here that's easy to miss: Cowork sessions come in two flavors — local (running on the user's own machine) and remote (triggered from web or mobile, running in Anthropic's cloud environment). According to the documentation, the Compliance API can retrieve transcripts for both, but a local session's conversation history is still fundamentally stored on the user's own computer. Enterprise admins can retrieve that content through the Compliance API, but there's no deletion endpoint available yet — meaning you can read a local session's records, but you can't yet centrally delete them through the API. That's a real limitation that remains even after this update.
Anthropic's announcement explicitly lists what this beta (now GA) excludes: Claude Code on the web, Claude Code accessed through the Claude Platform, and sessions run on Amazon Bedrock, Google Cloud's Vertex AI, or Microsoft Foundry. If your organization deploys Claude through any of those cloud platforms, this expanded Compliance API coverage doesn't reach you yet, and you're still relying on whatever monitoring approach you already had. That also means when assessing your organization's compliance posture, the question isn't simply "is Cowork covered now" in the abstract — it's whether your specific deployment actually falls within this expanded coverage.
Anthropic specifically notes that the new endpoints are additive: organizations already pulling data through the Compliance API see no change to what they already have, and organizations already exporting OpenTelemetry data can keep it running — the Compliance API works alongside it with no additional infrastructure required. That means if your organization already invested in building an OpenTelemetry streaming pipeline specifically to close the "Cowork has no audit trail" gap, that investment isn't wasted now that the Compliance API covers Cowork. The two can coexist: the Compliance API provides a more stable, structured per-session audit record, while OpenTelemetry continues serving real-time operational monitoring — their purposes were never fully overlapping to begin with.
If your organization previously restricted employees to only using regular Claude chat for regulated work, citing "Cowork has no audit trail" as the reason, this is a good moment to reassess whether that restriction still holds — it was valid before mid-August, and it no longer is. Before loosening that restriction, though, it's worth confirming a few things: whether your organization has the Compliance API enabled at all (if not, the new session endpoints need enabling through your existing Compliance Access Key); whether your actual deployment falls within this coverage (native Cowork desktop/web/mobile qualifies, but deployments through Bedrock, Vertex AI, or Microsoft Foundry don't yet); and whether the current lack of a deletion endpoint for local sessions matters for your organization's data retention policy. The claim that "Cowork has zero audit capability" is still circulating online and is now outdated — but outdated information tends to linger far longer than the actual change took to happen, so it's worth citing this specific update directly when discussing your compliance posture with colleagues or leadership, rather than relying on an analysis piece written a few months ago.