The Agentic Enterprise: Accountability and Containment When the Machine Decides

Agentic AI has moved from pilots to systems that touch real credit decisions, real client money and real compliance rules, and the institutions getting the governance right this year are setting the ceiling for the next decade. This session examines where accountability sits once an agent, not a person, initiates the action, across risk, audit and reporting lines — and then confronts the scenario most institutions would rather not discuss: what actually happens in the moments after an agent goes wrong.

An agent doesn't take a coffee break. When it fails, it can fail at machine speed, across every transaction it touches, before a human even notices. So the real test isn't whether agents fail, but how fast an institution can contain them when they do, and who is authorised to make that call. This discussion moves from the accountability question to the operational one: what real-time containment looks like in practice, where autonomous systems are most likely to break, and how to build the kill switches and safeguards that limit the blast radius before autonomy is extended any further.

Attendees leave with a sharper view of the controls that need to exist before autonomy is extended further. One of the key questions the panel will look to dive deeper into: Who is accountable when an autonomous agent makes a bad call, moves client money, or silently breaks a compliance rule?