{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/6f68ad9724174428b19db38926605d17\" frameborder=\"0\" width=\"1242\" height=\"931\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":931,"width":1242,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":931,"thumbnail_width":1242,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/6f68ad9724174428b19db38926605d17-cdf813fd40f5f061.gif","duration":149.805,"title":"Agent Patrol MVP Runtime Security Demo","description":"This Loom demonstrates Agent Patrol MVP as a runtime security layer that blocks malicious actions by rogue agents. The runtime wraps the agent, assigns an identity, and fetches an activity policy from the backend before any agent code runs, stopping commands outside the workspace, file access outside the workspace, and anomalous repeated calls. Each intercepted event is sent to the backend in real time, and the video shows a full threat log with enforcement outcome, per event risk score, and AO verdict from LLM analysis. It then tightens the policy from the dashboard, and on the next execution the same agent faces stricter product-wide blocking, with the updated policy snapshot attached to every new dashboard log entry."}