Research · curated 6 Aug 2026
ICML Poster Eyes-on-Me: Scalable RAG Poisoning through Transferable Attention-Steering Attractors
First reported icml.cc
Coverage timeline
Single-source research — first reported, latest, and curated coincide.
Why it matters
RAG poisoning that transfers across systems and steers model attention threatens the integrity of retrieval-augmented LLM deployments, a class defenders increasingly rely on for grounded outputs.
"Eyes-on-Me: Scalable RAG Poisoning through Transferable Attention-Steering Attractors," an ICML 2026 poster by Yen-Shan Chen and colleagues, presents a RAG data-poisoning technique that uses transferable attention-steering "attractors" to manipulate retrieval-augmented generation systems. The abstract is truncated, but the work is framed as a scalable, transferable poisoning attack against RAG pipelines.