OneBench
One Axis, No Brake: Self-Knowledge Limits the Filtering of Harmful Peer Conformity in LLMs | OneBench: AI Insights