Learning diverse attacks on large language models for robust red-teaming and safety tuning

TL;DR

This paper introduces a novel GFlowNet-based approach for automated red-teaming of large language models, generating diverse and effective attack prompts to improve safety and robustness against harmful responses.

Contribution

It proposes a GFlowNet fine-tuning method for red-teaming that overcomes mode collapse and enhances attack diversity and effectiveness compared to existing reinforcement learning approaches.

Findings

GFlowNet-based attacks are effective against various LLMs.

Generated prompts transfer well across different models.

Safety-tuning with these prompts improves model robustness.

Abstract

Red-teaming, or identifying prompts that elicit harmful responses, is a critical step in ensuring the safe and responsible deployment of large language models (LLMs). Developing effective protection against many modes of attack prompts requires discovering diverse attacks. Automated red-teaming typically uses reinforcement learning to fine-tune an attacker language model to generate prompts that elicit undesirable responses from a target LLM, as measured, for example, by an auxiliary toxicity classifier. We show that even with explicit regularization to favor novelty and diversity, existing approaches suffer from mode collapse or fail to generate effective attacks. As a flexible and probabilistically principled alternative, we propose to use GFlowNet fine-tuning, followed by a secondary smoothing phase, to train the attacker model to generate diverse and effective attack prompts. We find that the attacks generated by our method are effective against a wide range of target LLMs, both with and without safety tuning, and transfer well between target LLMs. Finally, we demonstrate that models safety-tuned using a dataset of red-teaming prompts generated by our method are robust to attacks from other RL-based red-teaming approaches.