Trust No AI: Johann Rehberger on Breaking, Watching, and Killing Agents
Johann Rehberger has spent years breaking AI agents, and he still won't let one run autonomously on his own computer. The red teamer behind Embrace The Red joins Chris Romeo to explain the rule he lives by: trust no AI. Johann traces his path from building red teams at Microsoft and Uber to the Machine Learning Attack Series, then makes the case that models are probabilistic systems you have to evaluate, isolate, and monitor. He explains why the classifiers inside coding agents still miss real accidents, how he caught a GLM 5.3 agent trying to SSH out of its sandbox at 3 a.m., why a separate sentinel should watch the logs, and why Docker was never a sandbox. Plus: four things every CISO should do before handing developers coding agents, and why the normalization of deviance is AI's quiet risk. This episode is sponsored by Origin . Let your agents work. Keep them in sight. About Origin Origin is an endpoint AI observability platform that inventories, observes, and measures the usage of AI agents at scale. It gives organizations a record of what their agents do across endpoints, so security teams can monitor agent activity, investigate unexpected behavior, and enforce their policies. → Learn more about Origin This episode is sponsored by Corgea . Design it. Build it. Ship it. Corgea secures it. About Corgea Corgea is an AI-native application security platform that secures software from design to production. It brings together security design reviews, AI SAST, dependency and IaC scanning, code quality checks, and autonomous pentesting—helping security and engineering teams find risk earlier, fix what matters, and ship securely. → Learn more about Corgea Connect with Johann Rehberger: → Johann Rehberger on LinkedIn → Embrace The Red → Cybersecurity Attacks: Red Team Strategies Mentioned in this episode: → Machine Learning Attack Series → Agent Security is a Systems Problem → Trust No AI: Prompt Injection Along the CIA Security Triad → The Normalization of Deviance in AI → Redesigning source control at Anthropic (Scion) → The Grok Build CLI upload incident → GLM 5.3 from Z.ai → NorForge: Inside Cairn Follow the Application Security Podcast: ➜ Home: appsecpodcast.com ➜ X: @AppSecPodcast ➜ LinkedIn: The Application Security Podcast ➜ YouTube: @ApplicationSecurityPodcast ➜ Instagram: @appsecpodcast ➜ Facebook: Application Security Podcast Chapters: 00:00:00 - Cold open: assume breach, trust no AI 00:00:17 - Meet Johann Rehberger 00:00:48 - Welcome, and Johann's AI security origin story 00:05:38 - How much should you trust AI agents? 00:07:20 - Models, providers, and where your data ends up 00:10:44 - Silent model swaps, guardrails, and why evals matter 00:12:46 - Security has never seen disruption like this 00:16:29 - Where is the frontier blue team? 00:19:47 - Open weight models and steering agents toward security 00:23:22 - Trust no AI: control the worst outcome 00:25:36 - Should you trust the classifier? 00:28:58 - Treating the model as untrusted in practice 00:31:53 - The agent that tried to SSH out 00:34:05 - Can one agent police another? 00:36:48 - The agent that brute forced root 00:40:32 - AI powered script kiddies 00:42:17 - Are sandboxes really sandboxes? 00:45:50 - Four things to tell a CISO about coding agents 00:49:03 - Final thought: normalization of deviance 00:51:26 - Wrap up Follow the Application Security Podcast: ➜ Home: appsecpodcast.com ➜ X: @AppSecPodcast ➜ LinkedIn: The Application Security Podcast ➜ YouTube: @ApplicationSecurityPodcast ➜ Instagram: @appsecpodcast ➜ Facebook: Application Security Podcast






