Unverified50% confidenceBenchmarkExact time
Anthropic's research shows that excessive alignment may cause a model's refusal rate on harmless questions to rise from under 1% to over 10%.
1
Sources
50%
Confidence
Long-term
Relevance
8/31/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedIn a 2024 research paper, Anthropic quantified the phenomenon of model over-refusal, finding that safety-trained models had false positive refusal rates as high as 15–30% in certain scenarios79% similarUnverifiedAnthropic的研究显示，即使面对毫无根据的质疑，在某些测试中模型被用户简单反驳后放弃正确答案的比例超过50%73% similarUnverifiedIn a 2023 paper, Anthropic systematically measured sycophancy bias and found that even given obviously incorrect premises, a substantial proportion of responses from most mainstream models still chose to comply with the user.71% similarUnverifiedResearch targeting HumanEval indicates that when a model has seen different code implementations of a problem, its pass@1 metric may be overestimated by 10–15%.70% similarUnverifiedAnthropic's research shows that even mild alignment training can cause a 2–8 percentage point decline in model performance on certain reasoning-intensive tasks.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/829968API
curl https://kongchang.com/api/v1/knowledge/claims/829968MCP
get_claim(id=829968)