Anthropic 在其商业产品（如 Claude for Work 和 API）中设定，默认情况下用户的输入与输出数据不会用于模型训练 1。然而，一旦用户通过“点赞”或“点踩”按钮主动提供反馈，包含内容与样式在内的整个相关对话将被存储于安全后端，最长保留期限可达 5 年 1。尽管 Anthropic 承诺在处理前会将反馈数据与用户 ID 及客户 ID 脱钩，但这些被留存的对话内容仍可能用于后续 AI 模型的训练 1。
在数据隐私边界方面，系统记录的反馈内容不包含连接器（例如 Google Drive）中的原始文件数据，但如果用户将这些外部内容直接复制粘贴进对话窗口，则会被一并纳入记录范围 1。针对企业层面的数据管控，Team 或 Enterprise 计划的所有者拥有权限，可通过调整“Rate chats”设置来禁用团队成员提交反馈的功能，从而进一步保护组织数据安全 1。
Anthropic has clarified its data handling policies for commercial products, noting that while user inputs and outputs are not used for model training by default, providing explicit feedback alters this condition 1. If users engage with features such as the "thumbs up" or "thumbs down" buttons to rate conversations, the associated dialogue content is stored on secure backends for a period of up to five years and may be utilized to train AI models 1.
The company specifies that this retention applies to the full context of the rated conversation, including both content and style elements 1. However, the feedback data itself does not include original content from external connectors like Google Drive unless it was directly copied into the chat interface 1. To address privacy concerns, Anthropic states that it decouples feedback data from user IDs and customer IDs before processing 1. Additionally, administrators within Team or Enterprise plans retain control over this feature, as they can disable the ability for members to submit feedback through the "Rate chats" settings in their organizational configuration 1.
评论
还没有评论，欢迎留下第一条。