Anthropic Ties Chat Feedback to Model Training — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic does not use inputs or outputs from commercial products—including Claude for Work, Anthropic API and Claude Gov—to train its models by default.
- Anthropic may use chats and coding sessions for training when a customer explicitly submits feedback or bugs, such as through Claude’s thumbs-up/down button, or otherwise allows data use.
- Anthropic stores the entire conversation attached to a thumbs report—including content, custom styles, conversation preferences and model settings—in a secured back end for up to 5 years.
- Anthropic de-links feedback from user and customer IDs before use and does not combine it with other Claude conversations.
- Google Drive raw content is excluded from feedback data, as is content from remote and local MCP servers; copied data may still be included if it appears directly in the Claude conversation.
- Claude for Work Primary Owners and Owners of Team or Enterprise plans can disable members’ feedback submissions through Organization settings > Data and Privacy > Rate chats.
- Claude consumer products—Free, Pro and Max—and Claude Code are outside the policy described, with those users directed to separate information.
Why it matters: For Claude for Work and Anthropic API customers, the default keeps commercial chats out of training, while explicit feedback gives Anthropic potential model-training material. The concrete trade-off is up to five years of conversation storage for each feedback report, with identifiers de-linked and raw connector files excluded unless copied into the chat.
Ask SkimNews
