The Authority to Stop: Who Can Halt the Agent When It Goes Wrong | Governed by Design #7

An AI agent is in production and starts doing the one thing it was built never to do. Who has the authority to stop it, right now, this afternoon?

We closed this series at five primitives. This episode is about the sixth, the one the regulation and the real-world deployments kept pointing at. Hosts Alex and Jaime introduce the Suspension Authority Ledger and why most governance answers the wrong question. You'll learn:

• Why "who approves the agent's decisions" and "who can pull it out of production" are different roles
• The two layers: oversight intervenes in a decision; suspension withdraws the system's authority
• Why the EU AI Act Omnibus extension moved the classification deadlines, but not the Article 26(5) duty to monitor and suspend, live from 2 August 2026
• The three fields of a Suspension Authority Ledger: a named authority (and backup), measurable triggers, and the notification chain
• The difference between a technical kill switch and the governance record of who's authorized to use it
• A worked example, a credit-triage agent halted in six minutes instead of a ten-minute blame spiral
• Why an untested ledger is a document, not a control, and how to rehearse it

This edition connects back to the Oversight Trigger Matrix (Edition 3) and the Accountability Canvas (Edition 4): the Canvas named owners for the agent's decisions; the Ledger names the owner for the agent's existence.

Key Insight: Oversight asks, should a person check this action? Suspension asks, should this system still be running? Two different questions, two different owners. Most governance only writes down the first.

Resources mentioned in this episode:📄 Full edition + the downloadable Suspension Authority Ledger template: themohamedadam.substack.com📰 LinkedIn newsletter version: Search "Automate & Elevate" on LinkedIn🔗 Website: aistreamlinehub.com - build a full profile at workspace.aistreamlinehub.com