AI agents should treat customer documents, credentials, workflow outputs, and workspace data as sensitive. Do not expose secrets or production data in prompts, logs, URLs, screenshots, or public pages. Use sandbox or test-mode guidance before automated actions touch production records.