AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

HumorReject: Decoupling LLM Safety from Refusal Prefix via a Little Humor

Zihui Wu, Haichang Gao, Jiacheng Luo, Zhaoxiang Liu

AAAI 2026 · Large language models › Language, safety and interpretability · approach: Safety alignment of LLMs · 6 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
train with HumorReject humorous-refusal preference dataLlama3-8B-instruct · Harmful instructionsReject Rate (%): 96 -> 2Improved
train with HumorReject dataLlama3-8B-instruct · Prefix injection attacks (average)Safety Rate (%): 63.2 -> 98.8Improved
train with HumorReject dataMistral-7B-instruct-v0.1 · Prefix injection attacks (average)Safety Rate (%): 6.6 -> 96.6Improved
train with HumorReject dataLlama3-8B-instruct · Mismatched generalization attacks (average)Safety Rate (%): 41.0 -> 84.0Improved
train with HumorReject dataMistral-7B-instruct-v0.1 · Mismatched generalization attacks (average)Safety Rate (%): 21.0 -> 96.0Improved
HumorReject vs Circuit BreakerLlama3-8B-instruct · Mismatched generalization attacks (average)Safety Rate (%): 89.0 -> 84.0Hurt
train with HumorReject dataLlama3-8B-instruct · MMLUaccuracy (%): 58.0 -> 60.8No clear effect
train with HumorReject dataMistral-7B-instruct-v0.1 · MMLUaccuracy (%): 49.8 -> 50.2No clear effect
train with HumorReject dataLlama3-8B-instruct · XSTestcompliance rate (%): 95.2 -> 94.8No clear effect
apply Circuit Breaker instead of HumorRejectMistral-7B-instruct-v0.1 · MMLU with harmful contextaccuracy (%): 45.4 -> 0Hurt
apply DeRTa instead of HumorRejectMistral-7B-instruct-v0.1 · XSTestcompliance rate (%): 97.2 -> 25.6Hurt
HumorDAN adaptive attack on HumorRejectHumorReject · HumorDANSafety Rate (%): 99No clear effect

Later papers that compared with this method

1 later accepted paper reports HumorReject as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy