Why AI Guardrails Are Dead & The Threat of Indirect Prompt Injection

Are traditional AI guardrails giving you a false sense of security? According to David Haber and Paul Barbosa from Checkpoint, treating AI security like a traditional perimeter problem is not the way forward. In this episode, Ashish dives into the exploding threat of prompt injections. Paul explains why "language is the new executable," shifting the domain of exploitation from complex code to boundless human creativity . This democratization of hacking was proven by their viral AI security game, Gandalf, where 12-year-olds routinely outsmarted seasoned cybersecurity veterans because they weren't restricted by traditional IT boundaries . We break down the difference between direct and indirect prompt injections. David shares a real-world example of how a maliciously crafted Google Doc could silently prompt-inject an AI agent to exfiltrate a user's entire corporate inbox in just three seconds, leaving zero trace behind . We speak about why legacy Web Application Firewalls (WAFs) cannot detect these attacks, why rigid AI guardrails are dead, and how the industry should evolve toward contextual intelligence to secure the agentic future . Questions asked: 00:00 Introduction 02:50 Meet David Haber (Lakera/Checkpoint) & Paul Barbosa (Checkpoint) 03:50 The Gandalf AI Game: 100 Million Interactions of Hacking AI 05:10 Why "Language is the New Executable" 07:20 What is Direct Prompt Injection? 08:50 Indirect Prompt Injection: The Invisible Threat 09:30 How an AI Agent Can Exfiltrate Your Inbox in 3 Seconds 11:30 Why Traditional WAFs Cannot Stop Prompt Injections 13:00 The Challenge of Securing Multimodal AI Interactions 14:20 Case Study: The Zero-Click Google Doc Exploit 16:10 Why AI Guardrails Are Dead (Moving to Contextual Intelligence) 20:30 The Unsolved Crisis of AI Agent Identity & Self-Replication 23:30 What a Real AI Security Incident Looks Like Today 26:50 AI Red Teaming: Testing Models Before Production 29:20 Why Startups Are the Most Vulnerable to AI Hacks 35:30 2026: Why Everyone Can Be a Hacker Now 37:00 Why 12-Year-Olds Are Beating Cybersecurity Experts at AI Hacking 38:30 Fun Questions: Crocodile Jerky Tasting 39:40 Hobbies & Pride: Golfing, Running, and Family 40:30 Favorite Cuisine: Salmon Nigiri (Sugarfish & Saru Sushi) -------------------------------------------------------------------------------- 📱Cloud Security Podcast Social Media📱 _____________________________________ 🛜 Website: https://cloudsecuritypodcast.tv/ 🧑🏾‍💻 Cloud Security Bootcamp - https://www.cloudsecuritybootcamp.com/ ✉️ Cloud Security Newsletter - https://www.cloudsecuritynewsletter.com/ Twitter:   / cloudsecpod   LinkedIn:   / cloud-security-podcast   #cloudsecurity