
AI Can Find Vulnerabilities Faster Than We Can Fix Them | Mahesh Shivnatri
Artificial intelligence is changing cybersecurity. But what happens when AI can discover vulnerabilities faster than organisations can fix them? In this episode of The Thought Atlas Podcast, I sit down with Mahesh Shivnatri, a technology transformation leader with more than 16 years of experience modernising enterprise systems across telecommunications, banking, insurance, and retail. A manager at Deloitte, Mahesh has led engineering teams working on security automation and is currently exploring AI-driven approaches to identifying vulnerabilities across large-scale technology environments. He is also an IEEE Senior Member, researcher, and peer reviewer whose work explores AI security, privacy, and enterprise transformation. We discuss the emergence of Frontier AI in cybersecurity, how major AI laboratories are approaching automated vulnerability discovery, and why the ability to identify security flaws is changing the way organisations think about defending their systems. Mahesh explains how AI-driven frameworks can identify vulnerabilities, develop potential fixes, test them in non-production environments, and submit them for human approval before deployment. We also explore Project Glasswing, Anthropic’s Claude Mythos Preview as discussed in the interview, and the growing importance of coordinated efforts between AI developers and cybersecurity defenders. One particularly important part of the conversation is virtual patching: a temporary protective measure that helps prevent attackers from exploiting a known vulnerability while organisations work towards a permanent fix. In this conversation, we explore: What Frontier AI means and how it is being applied to cybersecurity. How AI frameworks can identify vulnerabilities and develop potential patches. What OpenAI, Google, Anthropic, and Microsoft are doing in AI-driven security research, as described by Mahesh. What Project Glasswing is and why it matters. What the reported vulnerability discoveries associated with Project Glasswing reveal about software security. Why powerful AI models could change both offensive and defensive cybersecurity capabilities. How AI is making social engineering, phishing, voice cloning, and impersonation more sophisticated. The security risks associated with prompt injection, data leakage, and over-permissioned AI tools. What virtual patching means and how it protects vulnerable systems. Why virtual patching cannot replace permanent software fixes. Why faster vulnerability discovery creates new challenges for security teams. What organisations should prioritise as AI-driven cybersecurity continues to evolve. About the Guest Mahesh Shivnatri is a technology transformation leader with over 16 years of experience across enterprise technology and cybersecurity. He is a manager at Deloitte, an IEEE Senior Member, researcher, and peer reviewer. His work explores AI security, privacy, enterprise transformation, and automated approaches to identifying vulnerabilities in large-scale technology environments. Connect with Mahesh: LinkedIn: Search for Mahesh Shivnatri. Food for Thought: “Frontier AI is the way to go.” Mahesh’s closing reflection highlights the potential of Frontier AI to help identify and address vulnerabilities in the software systems that have become part of our everyday lives. About The Thought Atlas Podcast The Thought Atlas explores intellectual conversations spanning science, technology, physics, economics, psychology, and everyday human life. Website: https://joinpodmatch.com/thoughtatlas If you enjoyed this conversation, subscribe, leave a review, and share the episode with someone interested in artificial intelligence, cybersecurity, and the future of technology.