Unverified70% confidenceFactExact time
Cross-lingual jailbreaking exploits the asymmetry between a model's concept space and its language space, as the learning of safety-refusal behavior is highly language-specific.
2
Sources
70%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
Unverified大型语言模型在无约束空间中会倾向于创造性发散，而非工程所需的收敛执行81% similarUnverifiedSeveral widely used multilingual models exhibited embedding space collapse, i.e., representations of different inputs became highly similar, making them unsuitable for similarity-based clustering tasks.80% similarUnverifiedSafety-aligned language models have a one-dimensional direction in activation space associated with refusing harmful requests.79% similarUnverifiedCross-lingual transfer in multilingual models mainly relies on a shared multilingual representation space, which exhibits a significant alignment gap between language pairs with large differences in grammatical structure.79% similarUnverified大语言模型在二维空间推理上存在系统性短板，因训练数据以文本为主、缺乏空间关系监督信号78% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/533597API
curl https://kongchang.com/api/v1/knowledge/claims/533597MCP
get_claim(id=533597)