AI & ML
impact 16
Defusing the Trigger: Plug-and-Play Defense for Backdoored LLMs via Tail-Risk Intrinsic Geometric Smoothing
Defusing the Trigger: Plug-and-Play Defense for Backdoored LLMs via Tail-Risk Intrinsic Geometric Smoothing arXiv:2604.24162v1 Announce Type: new Abstract: Defending against backdoor attacks in large language models rem…
Why it matters
This signals a broader shift in defusing. The real question is whether trigger moves the needle for practitioners.