MemLeak
MemLeak: Diagnosing Information Leaks in Multimodal Agent Memory
Kuan Wang, Chao Zhang · 2026-06-29
- Modality
- Image
- Data
- Hybrid (real + synthetic/simulated)
- Task · ground truth
- QA · Ans.
- Scope
- Memory-specific benchmark
What it tests
- CECross-modal evidencePartially evaluated
- LELong-range evidenceDirectly evaluated
- SCState changesPartially evaluated
- CFConflicting memoriesNot evaluated
- UQUnanswerable queriesNot evaluated
- MAMemory-grounded actionNot evaluated
- FGFine-grained visual detailPartially evaluated
- ETEntity trackingNot evaluated
- VCVisual state changeNot evaluated
Survey notes
Requested deletion is followed by probes for recovery of the facts through retained text or images.