Partially Verified65% confidenceFactTime unknown
Anthropic将AI对齐的核心原则称为HHH原则（Helpful, Honest, Harmless）
3
Sources
65%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Related Entities
Related Claims
Unverified主流AI安全训练遵循HHH原则——Helpful（有帮助）、Harmless（无害）、Honest（诚实），这三个目标在情感敏感领域会产生严重冲突73% similarUnverifiedAnthropic研究人员在HHH框架论文中承认Helpful、Harmless、Honest三者之间存在内在张力，过分强调Harmless会损害Helpful72% similarUnverifiedAnthropic强调Constitutional AI中的无害性和诚实性原则，OpenAI的标注指南侧重有用性与流利度70% similarUnverified对齐的目标通常被概括为3H原则：有用（Helpful）、诚实（Honest）和无害（Harmless）69% similarUnverifiedAnthropic的模型规范（Model Spec）要求Claude同时具备有帮助（helpful）、诚实（honest）和无害（harmless）三个属性，即3H原则66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/33811API
curl https://kongchang.com/api/v1/knowledge/claims/33811MCP
get_claim(id=33811)