Automatic LLM Red Teaming

TL;DR

This paper introduces a novel reinforcement learning-based framework for automated red teaming of large language models, enabling the discovery of complex, multi-turn vulnerabilities through strategic, dynamic interactions.

Contribution

It formalizes red teaming as an MDP and employs hierarchical RL to generate multi-turn attack strategies, surpassing existing single-turn methods.

Findings

Achieves state-of-the-art vulnerability detection in LLMs.

Effectively uncovers subtle, multi-turn vulnerabilities.

Demonstrates the importance of dynamic, trajectory-based red teaming.

Abstract

Red teaming is critical for identifying vulnerabilities and building trust in current LLMs. However, current automated methods for Large Language Models (LLMs) rely on brittle prompt templates or single-turn attacks, failing to capture the complex, interactive nature of real-world adversarial dialogues. We propose a novel paradigm: training an AI to strategically `break' another AI. By formalizing red teaming as a Markov Decision Process (MDP) and employing a hierarchical Reinforcement Learning (RL) framework, we effectively address the inherent sparse reward and long-horizon challenges. Our generative agent learns coherent, multi-turn attack strategies through a fine-grained, token-level harm reward, enabling it to uncover subtle vulnerabilities missed by existing baselines. This approach sets a new state-of-the-art, fundamentally reframing LLM red teaming as a dynamic, trajectory-based process (rather than a one-step test) essential for robust AI deployment.