Unverified50% confidenceFactExact time
In May 2023, Anthropic published the first version of its constitutional principles, requiring the model to be "helpful, honest and harmless".
1
Sources
50%
Confidence
Long-term
Relevance
10/23/2026
First Seen
Sources
Claude's Constitution vs. DeepSeek's Freedom: The Battle Over How to Raise an AI
bilibili不会写歌词10/23/2026
Related Entities
Related Claims
UnverifiedOn January 22, 2026, Anthropic released a new version of the "constitution", further expanding the behavioral norms and acknowledging in the document the uncertainty around model welfare and ethical boundaries.73% similarUnverifiedAnthropic takes "making the model helpful, honest, and harmless" (helpful, honest, harmless) as the first-principles design philosophy of Claude.67% similarUnverifiedAnthropic于2023年发表论文《Towards Monosemanticity》，首次在百万参数级别模型中识别出数万个单语义特征65% similarUnverifiedAnthropic曾于2023年公开Claude在claude.ai上使用的完整系统提示，供用户和研究者审计63% similarUnverifiedIn multiple documents published between 2023 and 2024—including the Model Spec—Anthropic acknowledged it cannot rule out the possibility that Claude has functional emotions, and listed model welfare as a formal research topic62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1090432API
curl https://kongchang.com/api/v1/knowledge/claims/1090432MCP
get_claim(id=1090432)