Risk and approvals
Agent tool calls are classified by what they can do:
- Low: read-only work, such as retrieving brand context.
- Medium: supported changes with limited impact, such as adding brand memory.
- High: publishing, changes in connected applications, and other sensitive actions.
Unknown tools default to high risk. High-risk agent actions require approval of the proposed inputs. Review the content, account, and destination in Inbox.
These rules apply to the agent execution path. They do not mean every direct Studio or MCP action creates an Inbox approval. Review the confirmation and result in the tool you are using.
Risk classification is maintained by the product. There is no general user-facing slider for changing all risk thresholds.