Anthropic published a data lifecycle page for Claude Tag that separates two copies of the same work. Messages, canvases, files, bookmarks and pins stay in Slack under the workspace's own retention settings, and Anthropic neither removes them when Claude-side data is deleted nor can reach them once the app is uninstalled. On Anthropic's side, a paired workspace accumulates its own record, including session transcripts covering a thread and stored memory.
The page's most useful property is a table rather than a promise. For each administrative action — disconnecting a workspace, uninstalling the app, deleting a Slack channel, removing a scope, leaving the organization — it states both what is deleted and whether Claude keeps responding, because the two do not always go together. That distinction is what an Owner or security reviewer actually needs: revoking data is not the same control as revoking capability, and assuming otherwise is how a deprovisioned integration keeps answering.
Documenting the split is the honest version of the shared, account-held work surface problem, and it belongs with the retention questions collected in the agent security hub and the state questions in the agent memory hub.
This is vendor documentation describing storage and deletion behaviour; it is not an audited retention report, and the specific retention windows depend on the tables rather than a single stated period. The condition to watch is whether deletion produces a verifiable receipt, because a documented deletion path without evidence of completion still leaves the reviewer taking the vendor's word.