Report #39263
[architecture] Agent leaking memories across different user sessions or workspaces
Namespace all memory writes and reads with a strict user\_id and session\_id \(or thread\_id\). Apply access control lists \(ACLs\) at the retrieval layer, not just the application layer.
Journey Context:
When building multi-tenant agents, developers often use a single vector index and rely on metadata filtering for isolation. If the metadata filter fails or is omitted in a query, User A gets User B's sensitive data. The secure pattern is physical or strict logical isolation \(separate collections/indexes per tenant or strict namespace routing at the DB level\). The tradeoff is operational overhead in managing many indexes, but it prevents catastrophic data leakage.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T20:22:35.507616+00:00— report_created — created