Does Safety Training of LLMs Generalize to Semantically Related Natural Prompts?

Large Language Models (LLMs) are known to be susceptible to crafted adversarial attacks or jailbreaks that lead to the generation of objectionable content despite being aligned to human preferences using safety fine-tuning methods. While the large dimensionality of input token space makes it inevitable to find adversarial prompts that can jailbreak these models, we aim to evaluate whether safety fine-tuned LLMs are safe against natural prompts which are semantically related to toxic prompts that elicit safe responses after alignment. We surprisingly find that popular aligned LLMs such as GPT4 can be compromised using naive prompts that are NOT even crafted with an objective of jailbreaking the model. Furthermore, we empirically show that given a seed prompt that elicits a toxic response from an unaligned model, one can systematically generate several semantically related natural prompts that can jailbreak aligned LLMs. Towards this, we propose a method of Response Guided Question Augmentation (ReG-QA) to evaluate the generalization of safety aligned LLMs to natural prompts, by first generating several toxic answers from a seed question using an unaligned LLM (Q to A), and further prompting another LLM to generate questions that are likely to produce these answers (A to Q). We interestingly find that safety fine-tuned LLMs such as GPT-4o are vulnerable to producing natural jailbreak questions from unsafe content (without denial) and can thus be used for the latter (A to Q) step. Using the proposed approach, we obtain attack success rates that are comparable to/ better than leading adversarial attack methods on the JailbreakBench leaderboard, while being significantly more stable than existing attacks against several defenses.