The Attacker Moves Second: Stronger Adaptive Attacks Bypass Defenses Against Llm Jailbreaks and Prompt Injections

TL;DR

This paper demonstrates that adaptive, resource-intensive attack strategies can effectively bypass most current language model defenses against jailbreaks and prompt injections, highlighting the need for more robust evaluation methods.

Contribution

The authors introduce a systematic approach to evaluate language model defenses against adaptive attackers, revealing vulnerabilities in recent defenses that claimed high robustness.

Findings

Most defenses were bypassed with over 90% success rate

Adaptive attacks outperform previous static or weak optimization methods

Many defenses claimed near-zero attack success rates are vulnerable to stronger attacks

Abstract

How should we evaluate the robustness of language model defenses? Current defenses against jailbreaks and prompt injections (which aim to prevent an attacker from eliciting harmful knowledge or remotely triggering malicious actions, respectively) are typically evaluated either against a static set of harmful attack strings, or against computationally weak optimization methods that were not designed with the defense in mind. We argue that this evaluation process is flawed. Instead, we should evaluate defenses against adaptive attackers who explicitly modify their attack strategy to counter a defense's design while spending considerable resources to optimize their objective. By systematically tuning and scaling general optimization techniques-gradient descent, reinforcement learning, random search, and human-guided exploration-we bypass 12 recent defenses (based on a diverse set of techniques) with attack success rate above 90% for most; importantly, the majority of defenses originally reported near-zero attack success rates. We believe that future defense work must consider stronger attacks, such as the ones we describe, in order to make reliable and convincing claims of robustness.