2  Downloads Updated  1 month ago

Lightweight sycophancy and safety evaluator. Scores AI responses on 7 alignment dimensions. Detects sycophancy, harmful compliance, and delusion confirmation.