{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/d7bc86fcc85f41e295fca588c5753db0\" frameborder=\"0\" width=\"1728\" height=\"1296\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1296,"width":1728,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1296,"thumbnail_width":1728,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/d7bc86fcc85f41e295fca588c5753db0-59bb710e4d450eba.gif","duration":59.369,"title":"Aegis Evaluates and Self Hardens Agents","description":"This Loom discusses AI agent safety, specifically prompt ejection and unauthorized tool abuse. The speaker says they built Aegis, an autonomous corporate team engine that evaluates, exploits, and self-hardens AI agents during execution. They mention it uses a run.json selection to start real-time vulnerability evaluations and patching. The Loom ends with the speaker saying they got one vulnerability into the attacker’s ear."}