liveBook · Manning

Risk-based human oversight

As explained in AI Governance

Risk-based human oversight assigns human review or approval requirements according to the potential consequences of an action.

Assign review requirements by an action’s potential consequences, not by the agent’s reasoning. Classify tools in the policy engine or tool registry: read-only actions can run freely, reversible operations need lightweight confirmation, and high-consequence actions require approval on every call. section 4.8.8

Match the control to the action:

  • Use a confirmation dialog for non-agentic assistants. section 4.8.8
  • Require approval for each high-consequence tool call; calibrate thresholds so reviewers do not rubber-stamp requests. section 4.8.8
  • Prefer framework-native flows, such as LangGraph interrupts or AWS Bedrock human-in-the-loop options, over custom approval workflows. section 4.8.8

How AI Governance covers it

How deep it goes
Classifies tools by consequence and assigns execution, confirmation, or per-call approval controls.
Examples
Uses read-only tools, reversible operations, payment APIs, and record deletion to illustrate escalating controls.
What you can do after
Apply consequence-based approvals using LangGraph interrupts or AWS Bedrock human-in-the-loop options.
How it connects
Connects to policy engines, tool registries, LangGraph interrupts, and AWS Bedrock human-in-the-loop controls.

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.