AegisLLM: Scaling Agentic Systems for Self-Reflective Defense in LLM Security
This addresses security vulnerabilities in LLMs for users and developers, offering an adaptive runtime alternative to static defenses, though it appears incremental as it builds on existing agentic and prompt optimization methods.
The paper tackles the problem of adversarial attacks and information leakage in LLMs by introducing AegisLLM, a cooperative multi-agent defense system that scales agentic reasoning at test-time, achieving near-perfect unlearning with only 20 training examples and a 51% improvement on jailbreaking benchmarks.
We introduce AegisLLM, a cooperative multi-agent defense against adversarial attacks and information leakage. In AegisLLM, a structured workflow of autonomous agents - orchestrator, deflector, responder, and evaluator - collaborate to ensure safe and compliant LLM outputs, while self-improving over time through prompt optimization. We show that scaling agentic reasoning system at test-time - both by incorporating additional agent roles and by leveraging automated prompt optimization (such as DSPy)- substantially enhances robustness without compromising model utility. This test-time defense enables real-time adaptability to evolving attacks, without requiring model retraining. Comprehensive evaluations across key threat scenarios, including unlearning and jailbreaking, demonstrate the effectiveness of AegisLLM. On the WMDP unlearning benchmark, AegisLLM achieves near-perfect unlearning with only 20 training examples and fewer than 300 LM calls. For jailbreaking benchmarks, we achieve 51% improvement compared to the base model on StrongReject, with false refusal rates of only 7.9% on PHTest compared to 18-55% for comparable methods. Our results highlight the advantages of adaptive, agentic reasoning over static defenses, establishing AegisLLM as a strong runtime alternative to traditional approaches based on model modifications. Code is available at https://github.com/zikuicai/aegisllm