AI RESEARCH
Tatemae: Detecting Alignment Faking via Tool Selection in LLMs
arXiv CS.AI
•
ArXi:2604.26511v1 Announce Type: cross Alignment faking (AF) occurs when an LLM strategically complies with