<?xml version="1.0" encoding="UTF-8"?><oembed><type>video</type><version>1.0</version><html>&lt;iframe src=&quot;https://www.loom.com/embed/a6f69d1bc5bd4544a5017781a9858370&quot; frameborder=&quot;0&quot; width=&quot;1280&quot; height=&quot;960&quot; webkitallowfullscreen mozallowfullscreen allowfullscreen&gt;&lt;/iframe&gt;</html><height>960</height><width>1280</width><provider_name>Loom</provider_name><provider_url>https://www.loom.com</provider_url><thumbnail_height>960</thumbnail_height><thumbnail_width>1280</thumbnail_width><thumbnail_url>https://cdn.loom.com/sessions/thumbnails/a6f69d1bc5bd4544a5017781a9858370-5b11a00fd7ae64fd.gif</thumbnail_url><duration>563.04</duration><title>Guardian Nodes for AI Workflow Enforcement</title><description>An n8n AI agent that sends emails, with Guardian in front of it. Same agent, three outcomes: one email allowed automatically, one denied outright, one held for approval — where I open the payload, change it from a test message to a $400 refund, approve the edited version, and watch the audit log record that the executed payload was the edited one, not the original.</description></oembed>