Vulnerability  ·  2026-09-04

Cheshire Cat AI episodic memory endpoint leaks all users' conversation data via missing per-user filtering

VulnerabilityMedium impactGlobalCVE-2026-85093
Published Sept 3, 2026 (CVSS 6.5), this is a broken access-control / IDOR-style flaw in Cheshire Cat AI's conversational-memory retrieval API.
Cheshire Cat AI is an open-source conversational AI framework with persistent long-term memory; a missing tenant-isolation check on the memory-retrieval endpoint means any low-privilege authenticated user can mass-harvest every other user's private conversation history — a direct privacy and data-exposure failure in an LLM memory subsystem.
The GET /memory/collections/{collection_id}/points endpoint fails to apply per-user filtering when retrieving episodic memory points. An authenticated attacker holding only MEMORY:READ permission can paginate through the collection to retrieve all users' stored conversation messages and personal data, not just their own.
Cheshire Cat AI core (GET /memory/collections/{collection_id}/points endpoint)
Apply vendor fix (see cheshire-cat-ai/core GitHub repository for patched release).
secalerts.coNVD
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →