Cybersecurity's Biggest Blind Spot Is Your AI | Episode 4

In this episode of Samvada, Shubhanga Prasad sits down with Sanket Sarkar, founder of one of the few cybersecurity startups coming out of India that has genuinely scaled. What started as a conversation about trust in financial services quickly became one of the most sobering discussions we have had on this show. Five years ago there was no ChatGPT. Today AI agents learn on the fly, adapt when an attack fails, and craft entirely new exploits by themselves. The 1% of people who could execute a sophisticated cyberattack has grown to 30% and that number is only going up. But the deeper shift is this: you do not hack an LLM. You manipulate it. And that changes everything about how we think about security in an AI-first world. Guests: Sanket Sarkar - Founder, Zeron What we cover: 00:00 Intro 03:13 Same Tools, 10x The Attackers 07:22 LLMs Aren't Hacked, They're Manipulated 11:27 The End of Quality Assurance in AI 15:59 Why ISO and SOC Are No Longer Enough 23:10 The Gemini Jailbreak Story 33:41 Deepfakes, Fake Passports and The Identity Crisis 40:43 Why India Has No Global Cybersecurity Company Yet 46:15 The Decisions Are Still Human 48:07 The Problem That Could Break The Internet This is not a conversation about headlines or hype. It is about what is actually happening on the ground and what you need to know if you are building or buying technology today. Subscribe for more conversations with thinkers, operators, and investors shaping financial services and the world of AI. #VoiceAI #Samvada #ConversationalAI #LLM #AIStartups #Fintech #GaneshGopalan #AjayMa #Karate Ventures"