<?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/1a9e85585c474ac09814e4ff01759c4c&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/1a9e85585c474ac09814e4ff01759c4c-1c3a6e92a68eb2cd.gif</thumbnail_url><duration>300.095</duration><title>Orientation Gate Prevents Costly AI Actions</title><description>I’m showing a working demo of orientation gate, a control that evaluates proposed actions before execution. On the left, a system proposes an action, the gate returns allow, warn, or block, and on the right the system state updates in real time. Incident one, an AI agent deleted a production database, and the gate blocked it before any execution. Incident two, it warned on a destructive large scale cleanup, deleting over 500 gigabytes, to force review. There was no action requested from viewers.</description></oembed>