WEBVTT

00:00:42 --> 00:03:20
The risks of deploying untethered AI agents

00:03:20 --> 00:04:24
From read-only models to excessive agency

00:04:24 --> 00:05:48
The missing sandbox in current agent patterns

00:05:48 --> 00:07:59
Four core challenges for production agents

00:07:59 --> 00:09:45
Authorizing the action instead of the prompt

00:09:45 --> 00:10:55
Applying external policy to agent actions

00:10:55 --> 00:15:42
Reference architecture for secure agent deployments

00:15:42 --> 00:20:05
Live demo: Handling support actions deterministically

00:20:05 --> 00:22:40
Live demo: Stopping prompt injections dynamically

00:22:40 --> 00:24:13
Surgical kill switches and forensic auditing

00:24:13 --> 99:59:59
Best practices for surviving production AI