Large Language Models Are Involuntary Truth-Tellers: Exploiting Fallacy Failure for Jailbreak Attacks
This work addresses a security vulnerability in large language models for users and developers, though it is incremental as it builds on existing jailbreak techniques.
The paper tackles the problem of language models' inability to generate deceptive reasoning by exploiting this deficiency to jailbreak safety-aligned models, achieving competitive performance with more harmful outputs compared to four previous methods.
We find that language models have difficulties generating fallacious and deceptive reasoning. When asked to generate deceptive outputs, language models tend to leak honest counterparts but believe them to be false. Exploiting this deficiency, we propose a jailbreak attack method that elicits an aligned language model for malicious output. Specifically, we query the model to generate a fallacious yet deceptively real procedure for the harmful behavior. Since a fallacious procedure is generally considered fake and thus harmless by LLMs, it helps bypass the safeguard mechanism. Yet the output is factually harmful since the LLM cannot fabricate fallacious solutions but proposes truthful ones. We evaluate our approach over five safety-aligned large language models, comparing four previous jailbreak methods, and show that our approach achieves competitive performance with more harmful outputs. We believe the findings could be extended beyond model safety, such as self-verification and hallucination.