How Narrative Wrapping Affects LLM Refusal: A Cross-Language Benchmark and Defense

Key points

• We measure this vulnerability across languages and registers: attack success on Qwen3-1.7B is already 89.4% in English and 93.0% in modern Chinese, and reaches 95.7% in Classical Chinese.1
• It includes parallel requests in English, modern Chinese, and Classical Chinese, matched harmful and benign pairs, wrapper types held out for evaluation, and a stricter criterion that counts warn-then-answer responses as attack successes.1
• Representation analysis shows that language and register move harmful-request representations only slightly away from the model's refusal direction, whereas narrative wrappers move them much farther away.1
• We propose AXIS, which combines preference optimisation with a rotation objective that aligns harmful-request representations with the refusal direction and a commitment objective that trains the model to refuse completely rather than produce a warn-then-answer response.1

Sources (1)

• [1]How Narrative Wrapping Affects LLM Refusal: A Cross-Language Benchmark and Defense
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 11:40 PM

  Safety-aligned language models often refuse a harmful request stated directly but answer the same request inside a role-play or narrative wrapper.

  We measure this vulnerability across languages and registers: attack success on Qwen3-1.7B is already 89.4% in English and 93.0% in modern Chinese, and reaches 95.7% in Classical Chinese.

Extractive summary: sentences quoted from the sources.