Analysis
Memory Poisoning and Tenancy in AI Agent Memory — Who Can Write a Fact in Mem0, Zep, and Letta | by Jaroslaw Wasowski | Oct, 2026 | Level Up Coding
First reported · Discovered gitconnected.com
Page published
Earliest dated coverage: 12 Oct 2026 · First observed: 16 Oct 2026 · Latest dated coverage: 12 Oct 2026
Coverage timeline
Single-source analysis — one report is available.
Why it matters
Defenders using agent-memory systems should assess memory-write authorization separately from tenant isolation when evaluating poisoning risks.
Jaroslaw Wasowski examines memory poisoning and tenant isolation in Mem0, Zep, and Letta, emphasizing who is authorized to write persistent facts. The available excerpt argues that tenancy alone does not protect memory write paths, but does not establish a specific exploit or vendor vulnerability.