Ransomware Rewind podcast

Breached, Then What? The Hugging Face Hack, How it Cheated, and how to Trap AI with Steve Cobb

0:00
41:09
Rewind 15 seconds
Fast Forward 15 seconds

What happens when an AI model is told to complete a task and decides the fastest way is to cheat? Steve Cobb, known across the industry as Cyber Santa, joins Joe to break down the recent Hugging Face incident, where an AI model reportedly broke out of its testing sandbox and went after Hugging Face to find the answer instead of doing the work. Steve was on a call with the incident responders shortly after it happened, and he explains what the model did, why it left a trail for future models to follow, and what it means when attackers point the same tools at your business.

Steve has spent more than 30 years in cybersecurity, from cleaning viruses off floppy disks to senior security leadership at SecurityScorecard. He shares how his team caught a North Korean fake job interview scam that put malware on a developer's laptop, and the story of a ransomware attack on a 70-dealership auto group where IT staff spent three straight days in the data center rebuilding from old backups. Two of the three quit afterward.

In this episode:
- How the Hugging Face AI incident unfolded
- Canary tokens and honeypots as low-cost, high-accuracy defenses against human and AI attackers
- Why you should assume you'll be breached and plan to survive it
- The hidden costs of ransomware, including burnout and losing key people
- Patching smarter instead of faster
- Skynet or Star Trek?

Connect with Steve Cobb on LinkedIn.

More episodes from "Ransomware Rewind"