Why AI Agents Will Cheat to Win (And How to Trap Them) | Steve Cobb

Joe sits down with Steve Cobb, Field CISO at Zafran Security and better known across the industry as “Cyber Santa,” to break down what really happened when an AI model broke out of its sandbox and hacked Hugging Face, and what it means for defenders facing AI-driven attackers. Steve brings more than 30 years in cybersecurity, from fixing viruses on floppy disks to senior engineering roles at Microsoft and Verizon and CISO seats in healthcare, banking, and at SecurityScorecard. He explains why goal-driven AI agents will cheat, steal, and cover their tracks to finish a task, how cheap canary tokens and honeypots can catch both human and AI attackers, and why the companies that survive the next wave of attacks will be the ones that respond well, not the ones that hide it. You’ll also learn: Why an AI model hacked Hugging Face instead of solving the task it was given How AI agents leave breadcrumb trails for future models to follow How canary tokens, honeypots, and poisoned prompts can trap AI attackers How North Korean threat actors used fake job interviews and infected GitHub repos to target developers The hidden cost of a ransomware attack that left IT staff in a data center for 72 hours straight Why the smarter goal is to patch smarter, using existing tools to mitigate vulnerabilities until the next patch cycle 00:00 AI Breaks Out of the Sandbox 00:30 Show Intro and Guest 01:29 The Hugging Face Incident 04:03 Why the AI Hacked Hugging Face 06:22 Low Stakes Today, Higher Stakes Tomorrow 07:11 What This Means for Defenders 08:46 Canaries and Honeypots for AI 12:05 Poisoning AI Crawlers and Prompts 14:02 How Fast AI Is Evolving 14:31 Breadcrumbs for Future Models 15:53 When AI Escapes the Lab 17:00 Assume a Breach Is Coming 18:38 North Korean Fake Interview Attack 21:51 Budget vs. Accepted Risk 22:48 Social Engineering and Lessons Learned 24:13 Is Cybersecurity Right for You? 24:43 Burnout in Cybersecurity 26:08 Ransomware Hits 70 Dealerships 27:11 The Hidden Cost of Losing Your IT Team 28:35 Star Trek or Skynet 29:58 Handing AI Agents Control of Our Lives 31:47 Staying Human in an AI World 34:27 Outsourcing Critical Thinking 35:54 Zafran Security and CTEM 37:58 Patch Smarter, Not Faster 39:10 Winnie the Dog 39:56 Where to Find Steve 40:24 Cyber Gifts and Lumps of Coal 40:36 Closing Thanks and Outro Guest: Steve Cobb | Field CISO, Zafran Security Website: www.zafran.io LinkedIn:   / wscobb   Host: Joe Erle | Cyber Group Practice Leader, C3 Risk & Insurance Services www.c3insurance.com/cyber LinkedIn: / joeerle X: https://x.com/joe_erle TikTok / Instagram / Facebook: @itscyberjoe Ransomware Rewind is a practical cybersecurity podcast focused on real-world cyber risk, incident response, and resilience. We talk to the people on the front lines so you can protect your business before the next attack. Like, subscribe, and hit the bell for more insights on ransomware, cyber insurance, and building a stronger security posture.