liveBook · Manning

Approval gate

As explained in AI Governance

An approval gate is a control that requires authorization before an action proceeds.

An approval gate requires authorization before an agent commits an action. It is especially important when actions are irreversible or carry significant legal or financial consequences. section 2.6.2 section 5.4.2

Apply approval gates at action points where policy or risk requires authorization:

  • Set explicit limits, such as requiring approval before booking a more expensive hotel. section 2.6.2
  • Request human approval when a plan exceeds a configured limit, such as delaying watering beyond the permitted interval. section 2.6.2
  • Make human review meaningful by providing enough time, context, and authority to disagree; otherwise, review can mask automated decision-making. section 5.4.2
  • Record who acted, when, and under which policy, and constrain actions through authorization limits or system configuration. section 2.6.2

When law or risk requires human judgment, design the system as decision support rather than hidden automation. An agent that acts autonomously can cause harm before a person sees its flawed output. section 5.4.2

How AI Governance covers it

How deep it goes
Locates approval at action commitment and identifies risk triggers, but does not prescribe implementation mechanics.
Examples
Uses a hotel-price limit and a homeowner-approved watering delay to illustrate conditional authorization.
What you can do after
Evaluate whether agent actions need approval by checking risk, irreversibility, and legal or financial consequences.
How it connects
Ties approval gates to agentic AI, human-in-the-loop review, authorization limits, policy controls, and audit evidence.

Also explained in

Compare how 2 books explain approval gate →

Related concepts

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.