Anthropic Excludes Commercial API Data From Model Training by Default

Anthropic maintains separate data policies for its commercial and consumer products. Commercial offerings—including the API, Claude for Work, and government versions—do not use customer conversations for model training unless users explicitly opt in. This contrasts with consumer tiers like Claude Free and Pro, which follow different terms.

When users submit feedback through the built-in rating system, Anthropic retains the associated conversation for up to five years in secure storage. This feedback may be used for research, service improvement analysis, and model training, but Anthropic de-identifies the data by removing user and customer ID associations before leveraging it. Content pulled from external sources like Google Drive is excluded from feedback retention unless explicitly pasted into the chat.

Organizations on team and enterprise plans can further restrict data contribution by disabling feedback submission entirely through their administrative settings, giving them direct control over whether employees contribute to Anthropic’s training processes.