Jailbreaks as Inference-Time Alignment: A Framework for Understanding Safety Failures in LLMs.
James Beetham, Souradip Chakraborty, Mengdi Wang, Furong Huang, Amrit Singh Bedi, Mubarak Shah
Browse the full EACL paper archive.
James Beetham, Souradip Chakraborty, Mengdi Wang, Furong Huang, Amrit Singh Bedi, Mubarak Shah
Browse the full EACL paper archive.