Anthropic发布了关于Claude输出内容使用的政策说明。[1]虽然用户拥有Claude生成的输出内容，但根据该政策，用户不得使用这些输出来训练或开发竞争性AI模型，除非获得Anthropic的书面许可。[1]
Anthropic明确禁止用Claude的输出来训练通用聊天机器人或开放式文本生成模型。[1]公司采取这一限制的理由是防止安全控制丧失、无法对部署进行监控，以及防止服务基础设施被用于构建直接竞争者。[1]
相比之下，Anthropic允许用户使用Claude的输出来训练不与其竞争的专用模型，包括情感分析工具、内容分类系统、摘要工具、信息提取工具、语义搜索工具和异常检测工具。[1]
Anthropic has clarified its policy on how customers may use outputs generated by Claude, implementing restrictions that prevent the use of these outputs to train or develop competing artificial intelligence models [1]. Although users own the content that Claude generates, they are prohibited from leveraging it to train general-purpose chatbots or open-ended text generation systems without explicit written permission from Anthropic [1].
The company permits Claude outputs to be used for training specialized models that do not compete with Anthropic's offerings, including sentiment analysis tools, content classification systems, summarization tools, information extraction systems, semantic search applications, and anomaly detection tools [1]. However, the use of Claude's outputs to create general chatbots, open-form text generation models, or to reverse-engineer training methodologies is forbidden [1].
Anthropic's rationale for these restrictions centers on maintaining the integrity of its safety controls, preventing loss of oversight once models are deployed, and ensuring that its service infrastructure is not repurposed to build direct competitors [1].