AI is enabled only after the customer approves the feature/provider and the associated data flow. Instructions, selected excerpts and necessary context may leave Beetl for the approved provider; generated drafts/provenance may be saved in the workspace. The grounded-answer interaction table's narrower storage does not describe every AI feature. Customer content is not authorised for provider general-model training; abuse-monitoring and retention exceptions must be specifically disclosed from the actual contract. No zero-retention claim is made by a request-level store: false setting alone.

AI can make mistakes and create similar output for different customers. Check accuracy, permission, personal data and relevance before relying on or publishing output. Keep secrets and excluded data out of prompts. Customer administrators can disable enabled AI through the supported controls; deployment-wide disabling is available operationally. Disabling future use does not erase previous provider records automatically. The DPA and approved provider retention terms apply to existing records. No AI feature authorises bypassing access permissions or source restrictions.