Language-Switching Triggers Take a Latent Detour Through Language Models 事件

Language-Switching Triggers Take a Latent Detour Through Language Models arXiv:2605.18646v2 Announce Type: replace Abstract: Backdoor attacks on language models pose a growing security concern, yet the internal mechanisms by which a trigger sequence hijacks model computations remain poorly understood. We identify a circuit underlying a language-switching backdoor in an 8B-parameter autoregressive language model, where a three-word Latin trigger (nine tokens) redirects English output to French.