Bad Models, Good Security

What happens when AI agents start behaving in ways nobody intended? And how do you monitor systems that can make their own decisions? In this episode of the mnemonic Security Podcast, Robby is joined by Zack Korman, co-founder of Embroidery, a company focused on monitoring AI agent behavior. Together they explore the emerging security challenges of autonomous AI agents, from sandboxing and detecting suspicious behavior to using AI models to monitor other AI models. They revisit the Hugging Face incident, discuss why deliberately introducing misbehaving models through "chaos monkey" testing might actually improve security, and explore how falling token costs could change the way we build and secure AI systems. Finally, they tackle the bigger question: Should we focus on making AI models behave, or on building security that can handle them when they don't? Send us Fan Mail