作者位次 #7/12ACL 2026Poster· Natural Language Processing
Emergent Misalignment via In-Context Learning: Narrow in-context examples can produce broadly misaligned LLM s
Nikita Afonin · Nikita Andriianov · Vahagn Hovhannisyan · Nikhil Bageshpura · Kyle Liu · Kevin Zhu · Sunishchal Dev · Ashwinee Panda · Oleg Rogov · Elena Tutubalina · Alexander Panchenko · Mikhail Seleznyov
摘要
Recent work has shown that narrow finetuning can produce broadly misaligned LLMs, a phenomenon termed emergent misalignment (EM). While concerning, these findings were limited to finetuning and activation steering, leaving out in-context learning (ICL). We therefore ask: does EM emerge in ICL? We...