SEAS: Self-Evolving Adversarial Safety Optimization for Large Language Models

TL;DR

SEAS is a novel iterative framework that enhances large language model security by self-evolving adversarial prompts, significantly improving robustness and reducing manual testing reliance.

Contribution

Introduces a self-evolving adversarial safety framework that iteratively refines attack and defense models to improve LLM security and safety.

Findings

Target model security level comparable to GPT-4 after iterations

Red Team attack success rate significantly increased

Framework reduces manual testing efforts

Abstract

As large language models (LLMs) continue to advance in capability and influence, ensuring their security and preventing harmful outputs has become crucial. A promising approach to address these concerns involves training models to automatically generate adversarial prompts for red teaming. However, the evolving subtlety of vulnerabilities in LLMs challenges the effectiveness of current adversarial methods, which struggle to specifically target and explore the weaknesses of these models. To tackle these challenges, we introduce the $\mathbf{S}\text{elf-}\mathbf{E}\text{volving }\mathbf{A}\text{dversarial }\mathbf{S}\text{afety }\mathbf{(SEAS)}$ optimization framework, which enhances security by leveraging data generated by the model itself. SEAS operates through three iterative stages: Initialization, Attack, and Adversarial Optimization, refining both the Red Team and Target models to improve robustness and safety. This framework reduces reliance on manual testing and significantly enhances the security capabilities of LLMs. Our contributions include a novel adversarial framework, a comprehensive safety dataset, and after three iterations, the Target model achieves a security level comparable to GPT-4, while the Red Team model shows a marked increase in attack success rate (ASR) against advanced models. Our code and datasets are released at https://SEAS-LLM.github.io/.