Expired75% confidenceFactTime unknown
Anthropic uses an automated classifier (the LLM-as-a-Judge method) to evaluate Claude's degree of sycophancy across four dimensions: willingness to contradict, firmness of position, appropriateness of praise, and frankness.
5
Sources
75%
Confidence
Medium-term (~90 days)
Relevance
6/1/2026
First Seen
Valid until: 8/30/2026(expired)
Sources
Claude谄媚行为研究:灵性话题谄媚率高达38%,Anthropic揭示AI诚实度短板
rss
Related Entities
Cite This Claim
Stable URI
https://kongchang.com/claim/32876API
curl https://kongchang.com/api/v1/knowledge/claims/32876MCP
get_claim(id=32876)