safetyfactbearishFive open-weight LLMs show operationally useless near-constant confidence scores for self-auditing, while a proprietary model shows informative confidenceArtificial Intelligence02 Aug 2026http://arxiv.org/abs/2607.28317v1