Information-Theoretic Privacy Control for Sequential Multi-Agent LLM Systems
Status: Under review, 2026
This paper develops an information-theoretic framework for analyzing and controlling representation leakage in sequential multi-agent LLM systems. It characterizes safety–utility–privacy trade-offs and evaluates leakage behavior under adaptive adversarial conditions.
Recommended citation: Asif, S., & Mohammadi Amiri, M. (2026). "Information-Theoretic Privacy Control for Sequential Multi-Agent LLM Systems." Under review.
Download Paper