Who's Accountable When the Agent Acts? | Governed by Design #4

An agent does something unexpected. You know what happened. You don't know who's responsible.

In multi-agent systems, accountability doesn't just get complicated, it disappears into the handoff. In this episode, hosts Alex and Jaime introduce the Accountability Canvas: a pre-deployment tool for assigning four named owners before any agent goes live.

You'll learn:

• Why accountability diffuses in agentic systems, assumed by all, held by none
• What Accenture and Wharton mean by "intelligence may be scalable, but accountability is not"
• How OWASP ASI08 (Cascading Failures) makes attribution in multi-agent systems structurally difficult
• The four roles on the Accountability Canvas: Boundary Owner, Oversight Owner, Error Response Owner, Boundary Update Owner
• Why EU AI Act Article 26 requires a named natural person, not a team or committee
• How to implement canvas-before-deployment as a forcing function
• How the canvas connects directly to Decision Boundary Contracts (Edition 2) and the Oversight Spectrum (Edition 3)

Key Insight: Delegation doesn't transfer ownership. The team that deploys Agent A still owns what Agent C does, because they started the chain. The Accountability Canvas makes that explicit before the agent acts.

Resources mentioned in this episode:📄 Full edition with research + downloadable Accountability Mapping Canvas: themohamedadam.substack.com📰 LinkedIn newsletter version: Search "Automate & Elevate" on LinkedIn

Coming next week: Edition 5 "Audit Trails That Survive Scrutiny". What to log, how to structure it, and what regulators will actually ask for.