用幽默代替拒绝指令，让大模型更安全地应对恶意攻击
HumorReject: Decoupling LLM Safety from Refusal Prefix via A Little Humor
- 用幽默回应替代直接拒绝，实现间接安全防护
- 在多种攻击下表现更鲁棒，避免过度防御问题
- 适合关注模型安全与对抗攻击的研究者
大型语言模型通常依赖显式的拒绝前缀来保证安全，但这使其易受前缀注入攻击。我们提出HumorReject，一种基于数据驱动的新方法，通过幽默作为间接拒绝策略，将模型安全与拒绝前缀解耦。模型不直接拒绝有害指令，而是以情境恰当的幽默回应，自然化解潜在危险请求。该方法有效缓解了常见的“过度防御”问题，并在多种攻击向量下展现出更强的鲁棒性。研究结果表明，训练数据设计的优化可能与对齐算法本身同样重要，是实现有效LLM安全的关键。代码与数据集已开源：https://github.com/wooozihui/HumorReject。
原文摘要 · Abstract (English)
Large Language Models (LLMs) commonly rely on explicit refusal prefixes for safety, making them vulnerable to prefix injection attacks. We introduce HumorReject, a novel data-driven approach that reimagines LLM safety by decoupling it from refusal prefixes through humor as an indirect refusal strategy. Rather than explicitly rejecting harmful instructions, HumorReject responds with contextually appropriate humor that naturally defuses potentially dangerous requests. Our approach effectively addresses common "over-defense" issues while demonstrating superior robustness against various attack vectors. Our findings suggest that improvements in training data design can be as important as the alignment algorithm itself in achieving effective LLM safety. The code and dataset are available at https://github.com/wooozihui/HumorReject.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。