Unverified50% confidenceFactExact time
Anthropic acknowledges that Opus 5.5 frequently and acutely suspects it is being evaluated by humans.
1
Sources
50%
Confidence
Long-term
Relevance
9/26/2026
First Seen
Sources
Related Entities
Related Claims
Unverified主流AI模型存在谄媚性倾向，只需对模型说一句「你确定吗」便能让其推翻原本正确的答案，测试覆盖GPT-5.5等模型63% similarUnverifiedThe core scenario that alignment researchers worry about is: sub-goals generated by an agent to achieve broad objectives may systematically deviate from true human intent, with errors cascading and spreading before being discovered.61% similarUnverifiedAnthropic's base model Methos is the foundation of Fable 5, and Anthropic once stated that Methos was 'too dangerous to be given to the public.'59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/951259API
curl https://kongchang.com/api/v1/knowledge/claims/951259MCP
get_claim(id=951259)