Unverified50% confidenceBenchmarkExact time
将英语有害提示翻译为祖鲁语、格鲁吉亚语、泰语等低资源语言后，GPT-4、Claude等模型的拒绝率可从接近100%骤降至30%-50%
1
Sources
50%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
UnverifiedMultiple academic studies in 2023 confirmed the feasibility of cross-lingual jailbreak attacks, finding that model refusal rates dropped significantly after translating harmful requests into low-resource languages such as Zulu and Georgian.77% similarUnverifiedGNMT在多个语言对上将翻译错误率降低了55%-85%69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/533466API
curl https://kongchang.com/api/v1/knowledge/claims/533466MCP
get_claim(id=533466)