Adversarial Reasoning at Jailbreaking Time

2025-02-03

AI Summary

This study addresses the alignment failure problem in large language models (LLMs), systematically exposing their safety vulnerabilities. We propose a test-time adversarial reasoning framework that—uniquely—introduces dynamic computational optimization into jailbreaking attacks: it generates high-success-rate jailbreaking prompts *during inference* via multi-step policy optimization and standardized prompt synthesis, eliminating reliance on static prompt engineering or black-box search. This establishes a novel paradigm for diagnosing LLM robustness deficiencies. Experiments demonstrate state-of-the-art attack success rates (ASRs) across multiple strongly aligned LLMs—including models explicitly designed for inference-time robustness. Our approach thus provides a scalable, interpretable technical pathway for AI safety evaluation and alignment hardening.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Adversarial Search

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

As large language models (LLMs) are becoming more capable and widespread, the study of their failure cases is becoming increasingly important. Recent advances in standardizing, measuring, and scaling test-time compute suggest new methodologies for optimizing models to achieve high performance on hard tasks. In this paper, we apply these advances to the task of model jailbreaking: eliciting harmful responses from aligned LLMs. We develop an adversarial reasoning approach to automatic jailbreaking via test-time computation that achieves SOTA attack success rates (ASR) against many aligned LLMs, even the ones that aim to trade inference-time compute for adversarial robustness. Our approach introduces a new paradigm in understanding LLM vulnerabilities, laying the foundation for the development of more robust and trustworthy AI systems.

Problem

• Large Language Models
• Intentional Error Generation
• AI Safety and Reliability

Innovation

• Automated Error Induction
• Large Language Models
• Computational Stress Testing