When Poison Fails After Retrieval: Revisiting Corpus Poisoning under Chunking and Reranking Pipelines

This collection of arXiv papers highlights several security- and privacy-relevant advances and risks across ML, crypto, and network covert channels. Key findings: (1) Retrieval-Augmented Generation (RAG) corpus poisoning is much less effective in realistic pipelines with chunking and reranking unless attacks are chunk-aware; the proposed CRCP method restores robust poisoning by optimizing for chunk-boundary and reranker consistency. (2) JailbreakOPT is a tool-assisted iterative framework that substantially improves jailbreak success rates against LLMs by composing and reusing atomic jailbreaks