作者位次 #3/7ACL 2026Findings· Findings of ACL
Privacy Risks of Intermediate Representations: Attribute Inference in Distributed LLM Inference
Yang Lyu · Jin Cao · Yang Xiao · Zhe Sun · Ben Niu · Fenghua Li · Hui Li
摘要
Distributed LLM inference avoids sending raw inputs by transmitting intermediate hidden states, a practice widely assumed to preserve privacy. We challenge this assumption and demonstrate that intermediate representations alone are sufficient to leak sensitive user attributes. This setting poses...