AI security guardrails exhibit uneven efficacy across non-English languages

AI safety mechanisms frequently fail to maintain consistent protection against jailbreaking and malicious prompts when operating in languages other than English. This linguistic disparity creates a significant attack surface for global deployments, as security testing often prioritizes English-language datasets.