# RLAIF

Constitutional AI Training Attacks

Attacking Constitutional AI and RLAIF training pipelines by manipulating the constitutional principles, critique models, or self-improvement loops.