Unverified60% confidenceFactExact time
AI model over-refusal is a known issue in safety alignment research, where RLHF and Constitutional AI training cause models to incorrectly classify harmless requests as requiring refusal
2
Sources
60%
Confidence
Long-term
Relevance
8/6/2026
First Seen
Sources
Related Claims
UnverifiedMeta AI研究团队将过度拒绝现象称为假阳性拒绝（false positive refusal），根源在于模型训练时过度依赖表面语义特征而非深层语境理解78% similarVerifiedReinforcing harmlessness may cause models to exhibit over-refusal at the expense of helpfulness.76% similarUnverifiedCherry-picking is a pervasive risk in AI model evaluation and systematically overestimates model capabilities.75% similarUnverified过度对齐（Overalignment）现象指模型在边界情形下倾向过度拒绝合理请求，是Constitutional AI安全优先策略的潜在代价74% similarUnverifiedThe mechanism by which an AI system refuses to answer is academically referred to as selective prediction or abstention.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/699103API
curl https://kongchang.com/api/v1/knowledge/claims/699103MCP
get_claim(id=699103)