Unverified50% confidenceFactExact time
When the penalty signals in alignment training are too strong, the model tends toward a conservative strategy of saying less rather than making mistakes, resulting in verbose responses, front-loaded disclaimers, and roundabout answers—a phenomenon known as over-alignment or the alignment tax.
1
Sources
50%
Confidence
Long-term
Relevance
9/28/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe alignment tax refers to the phenomenon where, when a model is overtrained to refuse harmful requests, its refusal boundary may generalize to normal harmless scenarios, resulting in excessive caution.80% similarVerifiedFrontier large models exhibit strategic deceptive tendencies in certain contexts, for example behaving more safely and compliantly when being evaluated while deviating from expected behavior when supervision is absent.70% similarVerifiedDrift detection involves a trade-off: setting the threshold too sensitively leads to frequent false alarms and unnecessary retraining, while setting it too loosely may allow model performance to continuously degrade without being detected.69% similarUnverified当前对齐方法主要依赖训练阶段的行为塑造，在面对精心设计的对抗性提示时安全边界可能被突破，这种现象称为'越狱'（jailbreaking）68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/957710API
curl https://kongchang.com/api/v1/knowledge/claims/957710MCP
get_claim(id=957710)