video1.0<iframe src="https://www.loom.com/embed/f50a62a2153b4a9882f690547756a41d" frameborder="0" width="1728" height="1296" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>12961728Loomhttps://www.loom.com12961728https://cdn.loom.com/sessions/thumbnails/f50a62a2153b4a9882f690547756a41d-384ba0158aeef59c.gif142.771This Loom demonstrates how Maya’s AI agent can be attacked through her emails, calendar invites, and documents, and shows defenses that prevent misuse. Alex presents a dashboard of four betrayal scenarios, including versioned agents using Memorable, QM, and G Brain, and explains that authorization and built-in security help block harmful actions. During live testing, he repeatedly attacks by sending an email pretending to be her billing company to update her invoice and by using Memorable via a doc instruction, but the defenses successfully detect and stop the attempts, including cases where the process proceeds only if memory is clean and contacts are not spammed. He concludes that these threats would be highly harmful if not defended, especially since Maya is a founder who relies on the agent to manage tasks.