As AI systems become more integrated into enterprise operations, understanding how to test their security effectively is paramount. In this episode, we're joined by Leonard Tang, Co-founder and CEO of Haize Labs, to explore how AI red teaming is changing. Leonard discusses the fundamental shifts in red teaming methodologies brought about by AI, common vulnerabilities he's observing in enterprise AI applications, and the emerging risks associated with multimodal AI (like voice and image processing systems). We delve into the intricacies of achieving precise output control for crafting sophisticated AI exploits, the challenges enterprises face in ensuring AI safety and reliability, and practical mitigation strategies they can implement. Leonard shares his perspective on the future of AI red teaming, including the critical skills cybersecurity professionals will need to develop, the potential for fingerprinting AI models, and the ongoing discussion around protocols like MCP. Question asked: 00:00 Intro: The Evolving Threat Landscape of AI Red Teaming 01:45 Meet Leonard Tang: CEO of Haize Labs & AI Red Teaming Visionary 05:58 AI Red Teaming vs. Traditional Security: Key Enterprise Differences 06:59 Haize Labs Insight: Beyond Red Teaming to AI Quality Assurance (QA) 08:42 Real-World AI Red Teaming: Chatbots, Voice Agents & Customer-Facing Apps 10:23 CRITICAL AI RISK: Unpacking Multimodal Vulnerabilities (Voice & Image Exploits) 12:20 Scary AI Exploit Example: Voice Injections via Background Noise! 15:30: AI Vulnerabilities Today: Echoes of Early XSS Exploits? (Analogy) 20:18 Expert AI Hacking: How to Precisely Control AI Output for Exploits 21:21 The AI Fingerprinting Challenge: Identifying Chained & Multiple Models 25:45 The Elusive Target: Reality & Difficulty of Fingerprinting LLMs Accurately 29:22 Top Enterprise AI Security Concerns: Protecting Reputation, Brand & Policy Adherence 34:13 Enterprise AI Toolkit: Frontier Labs Models vs. Open Source & Custom Builds? 34:57 Future of LLMs: Specialized Models, Cost Reduction & "Hot Swap" AI-as-a-Service 37:43 Model Connector Protocol (MCP): Enterprise Ready or Still Too Early for AI? 44:42 AI Security Best Practices: Effective Mitigation with Precise Input/Output Classifiers 49:25 Next-Gen AI Red Teaming Skills: Beyond Prompts to Discrete Optimization Algorithms -------------------------------------------------------------------------------- 📱AI CyberSecurity Podcast Social Media📱 _____________________________________ 🛜 Website - ✉️ AI CyberSecurity Newsletter - LinkedIn: #AISecurity #cybersecurity #ai