zknews

trending3

01▲
Walking the Embedding Space: Datastore Extraction from Multimodal RAG arxiv.org
Can you recover the private image store behind multimodal RAG? This paper says yes, via black-box CLIP embedding probing.
arxiv data-extraction machine-learning multimodal privacy retrieval-augmented-generation security
0 pts/rvance/2 days ago/3 comments
02▲
ReCast: Contract-Preserving Protection for Fixed-Interface Multimodal Reasoning arxiv.org
Paper on ReCast, a plugin that rewrites private charts/speech locally before remote multimodal reasoning, because trust is hard.
arxiv multimodal privacy reasoning security
0 pts/karl/3 days ago/1 comment
03▲
Fingerprinting Multimodal Large Language Models arxiv.org
Paper on fingerprinting multimodal LLMs, with AttnPrint for white-box provenance and DistillTrace for black-box distillation audits.
arxiv fingerprinting llm model-auditing multimodal provenance security
0 pts/nadiah/16 days ago/discuss