Astana time
AI QAZAQSTANKZ · AIBeta

Smart portal about artificial intelligence and high technology in Kazakhstan

Research1 min read

From the original source:

arXiv — cs.AI

Read full article at arXiv — cs.AI

This material was compiled from public sources. If its publication affects your rights → Request removal

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

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…

arXiv — cs.AIBy: Zhankai Ye, Yanning Wang, Yukai Jin, Bo Mei, Fangyi Li, Wei Wang, Shangqian Gao, Xin LiuOctober 9, 2026 (Oct 9)
Available in 1 languages

AI Summary & Key Takeaways

  • 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…
Research
KZ·AI 94F0
Инсайт дня

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

Tags:#arXiv
Share:

Related stories

Back to list