MemLeak benchmark shows retained images enable 12% recovery of deleted facts in multimodal agents (reduced to 2% with content-aware deletion), with 47% of image leaks not text-recoverable.
Rethinking visual privacy: A compositional privacy risk framework for severity assessment with vlms
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
years
2026 2verdicts
UNVERDICTED 2roles
background 1polarities
background 1representative citing papers
Vision-language models exhibit perceptual fragility and fail to consistently respect privacy constraints when operating in simulated physical environments, with performance declining in cluttered scenes and under conflicting commands.
citing papers explorer
-
MemLeak: Diagnosing Information Leaks in Multimodal Agent Memory
MemLeak benchmark shows retained images enable 12% recovery of deleted facts in multimodal agents (reduced to 2% with content-aware deletion), with 47% of image leaks not text-recoverable.