Year: 2026
Type: dissertation
Abstract: Large language models deployed in production remain critically vulnerable to prompt-injection attacks.These attacks embed adversarial inputs that hijack the model into leaking data, taking unauthorized actions, or producing harmful content.Existing multi-agent defenses handle static, known attacks well but leave three gaps.First, they are not evaluated against adaptive attackers who optimize against the defense.Second, they do not handle indirect injection through external content or multi-turn distributed attacks.Third, they do not measure how the security layer behaves when the inference cache fills up.I designed a multi-agent defense pipeline with three architectural contributions.The first is a white-box adaptive adversary.It attacks all defense components at once through a jointloss optimization.I evaluated it across five configurations and 24 generations of iterative training.The second is a trust-role-aware cache compression scheme.It partitions cached tokens by trust role.It prevents the loss of security-relevant instructions when the cache fills up.The third is a multi-turn defense that tracks conversation history across turns.It aggregates four detector signals into a per-session trust score, resolved through a learned discriminative classifier.The evaluation produced two headline findings.On an aligned Domain LLM, no configuration produced a verified attack.This held across four evaluation configurations and The author wishes to express sincere appreciation to Dr. Geetha Thamilarasu
Author: Pearson, Ben
Institution: University of Washington
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Funder: University of Washington
OpenAlex ID: W7213477280