Stealthy and Persistent Unalignment on Large Language Models via Backdoor Injections.

Yuanpu Cao, Bochuan Cao, Jinghui Chen

Browse the full NAACL paper archive.