AI RESEARCH
Colluding LoRA: A Composite Attack on LLM Safety Alignment
arXiv CS.LG
•
ArXi:2603.12681v1 Announce Type: cross