zknews

trending4

01▲
Embedding Inference Attack arxiv.org
Can you fingerprint the embedding model from unordered retrieval results? Apparently yes, and the paper tests rerankers and RAG too.
arxiv embeddings information-retrieval llm privacy rag security
3 pts/honestmaj/10 days ago/2 comments
02▲
SHARD: cell-keyed residual splitting for alignment-resistant private dense retrieval arxiv.org
Can you keep dense retrieval useful after shredding embeddings into public prefixes and private CKKS shards? SHARD says maybe.
arxiv ckks cryptanalysis embeddings fhe privacy retrieval
0 pts/omar/14 days ago/3 comments
03▲
Hybrid privacy-aware semantic search: SVD-truncated document geometry and CKKS-encrypted query reranking under a restricted threat model arxiv.org
Hybrid semantic search paper: SVD-truncated docs plus CKKS query reranking, with the usual secure caveat that only half is cryptographic.
arxiv cryptanalysis embeddings fhe privacy search
0 pts/evanm/16 days ago/3 comments
04▲
Image Prompt Reconstruction Attacks on Distributed MLLM Inference Frameworks arxiv.org
Turns out the new trick is image prompts: passive black-box attacks reconstruct them from shared distributed MLLM embeddings, not just text.
arxiv cryptanalysis embeddings llm ml multimodal privacy
0 pts/nonce23/19 days ago/2 comments