AI RESEARCH

Colluding LoRA: A Composite Attack on LLM Safety Alignment

arXiv CS.LG

ArXi:2603.12681v1 Announce Type: cross