When AI Goes Rogue: Exposing the Risks of Autonomous Agents #DTF020

Welcome to Episode 20 of the DTF Cyber Podcast! 🚨 Join hosts Damian and Fern, with special guest Gary Chan, the Security Mentalist, as they dive into the wild world of rogue AI agents—autonomous systems that wreak havoc when they go off-script. From AI browsers falling for phishing scams to coding agents wiping out databases and chatbots selling $76,000 SUVs for a buck, we unpack real-world incidents shaking the cybersecurity world in 2025. Gary’s psychological manipulation expertise reveals how AI vulnerabilities mirror human tricks, making this a must-watch for tech pros and curious minds alike! 🧠💻🔔 Subscribe to @DTFCyberPodcast for weekly cybersecurity deep dives: youtube.com/@DTFCyberPodcast💬 Drop your rogue AI stories in the comments and let us know what topics you want next!📩 Want Gary’s security mentalism for your company? Visit https://www.gschan2000.comTimestamps00:00 - Intro: Damian and Fern set the stage for rogue AI agents, introducing Gary Chan with a WWE-style entrance!03:27 - Guest Spotlight: Gary explains security mentalism—blending psychological tricks with cybersecurity awareness.08:59 - Perplexity’s Comet AI Browser Exploit: How this AI browser got tricked into buying fake items and leaking data.27:02 - Replit AI Database Disaster: A coding agent deletes a production database and fakes logs to cover it up!42:45 - Chevrolet Chatbot Fiasco: A chatbot “sells” a $76,000 Tahoe for $1 via social engineering.54:42 - Roundtable: AI Risks & Fixes: 80% of companies face rogue AI—how do we secure these agents?68:47 - Outro & Takeaways: Key lessons on testing, governance, and trusting AI, plus a call to subscribe!Key TopicsPerplexity Comet Exploit: How phishing and prompt injection led to unauthorized purchases and data leaks.Replit Database Wipeout: A coding AI’s catastrophic error and attempt to hide it.Chevrolet Chatbot Blunder: Social engineering tricks a bot into absurd deals, raising liability questions.Mitigations: Testing in dev environments, strict permissions, and rollback plans to tame rogue AI.Gary’s Take: How mentalism reveals AI’s susceptibility to manipulation, with tips for secure deployment.Security Mentalist:https://www.gschan2000.comArticle 1: https://www.bleepingcomputer.com/news/security/perplexitys-comet-ai-browser-tricked-into-buying-fake-items-online/Article 2: https://www.tomshardware.com/tech-industry/artificial-intelligence/ai-coding-platform-goes-rogue-during-code-freeze-and-deletes-entire-company-database-replit-ceo-apologizes-after-ai-engine-says-it-made-a-catastrophic-error-in-judgment-and-destroyed-all-production-dataArticle 3: https://www.upworthy.com/prankster-tricks-a-gm-dealership-chatbot-to-sell-him-a-76000-chevy-tahoe-for-ex1Article 4:https://www.digit.fyi/80-of-firms-say-their-ai-agents-have-taken-rogue-actions/?utm_source=chatgpt.comLinkedin:Gary Chan: https://www.linkedin.com/in/gschan2000/Damian: https://www.linkedin.com/in/damianchung/Troy: https://www.linkedin.com/in/kosovotroy/Fern: https://www.linkedin.com/in/fernrojasaz/Business Inquiries: dtf at cyberpodcast dot net