arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2606.01196cs.CLcs.AI
收藏

低资源安全失败是行动失败,而非表征失败

Low-Resource Safety Failures Are Action Failures, Not Representation Failures

Mohamed bin Zayed University of Artificial Intelligence(莫扎伊德大学人工智能学院)

Rashad Aziz, Ikhlasul Akmal Hanif, Fajri Koto

AI总结:

本文发现低资源语言的安全对齐失败源于决策校准问题而非表征缺失,通过重校准高资源门控(低秩逻辑回归+阈值重置)显著提升拒绝选择性。

摘要:

在高资源语言中学习的安全对齐在低资源语言中迁移效果不佳。模型能拒绝英文有害提示,但当相同提示翻译成斯瓦希里语或缅甸语时则无法拒绝。自适应引导方法如AdaSteer和CAST在跨语言中继承了这一失败。我们诊断了迁移失败的原因。在Qwen2.5-7B、Gemma-2-9B和Llama-3.1-8B模型上,针对23种语言,从高资源激活中提取的有害方向几乎能像高资源提示一样线性分离低资源有害与无害提示。相关表征存在。然而,有害拒绝率从87.9%下降到43.9%。模型未能将表征转化为拒绝。未能迁移的是安全决策的校准,而非底层表征。我们利用这一点,通过重校准而非重新训练高资源门控:一个低秩逻辑回归读出器,其决策阈值使用每类仅1到4个目标语言示例重置。该门控在拒绝引导和有害方向消融之间路由,将平均拒绝选择性(Δ = 有害 − 无害拒绝)从最强自适应基线的33.6显著提高到54.5,同时保持MMLU效用。这些结果表明,一些低资源安全失败可以通过重校准现有表征而非学习新表征来修复。我们的代码已发布:https://github.com/rashadaziz/low-resource-safety。

英文摘要:

Language models often answer harmful requests in low-resource languages (LRLs) that they refuse in high-resource languages (HRLs). Across three instruction-tuned models and 23 languages, harmful refusal falls from 87.9% in HRLs to 43.9% in LRLs, while harmless refusal remains low. A common explanation is that models represent harmfulness weakly in LRLs. We test whether harmfulness is instead represented but does not reliably produce refusal. Across three models, a harmfulness direction learned from HRL activations still separates harmful from harmless LRL prompts, showing that complete absence of harmfulness information cannot explain many failures. However, harmfulness scores shift downward for LRL prompts, making harmful prompts less likely to reach the range associated with refusal. Motivated by this shift, we train a low-rank logistic classifier on HRL activations and calibrate its threshold with a few target-language examples. During generation, the classifier conditionally adds or ablates the HRL harmfulness direction. With the same HRL data and 32 target-language examples per class, CAST remains limited by low harmful refusal and AdaSteer by high harmless refusal, yielding mean refusal selectivity ($Δ$ = harmful - harmless refusal) of 33.6 and 6.8, respectively. Our intervention reaches 54.5 while preserving MMLU utility. HRL-only calibration improves selectivity for Qwen and Gemma, whereas Llama benefits from target-language calibration. These results show that recalibrating existing representations can offer a training-free method for repairing low-resource safety failures.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑