Why AI Agents
Go Off the Rails
AI agents don’t fail like normal software. Here’s what that means for defenders.
An agent doesn’t need to be hacked
to do the wrong thing.
AI agents don’t fail like normal software. They take the wrong action with total confidence, chain small mistakes into big ones, and get talked into things by the data they read.
Etay Maor has spent his career studying how attackers think, from fraud rings to threat intelligence. He joins Dan Barahona for a live conversation on why agents go off the rails, and what security teams can do about it before they’re running everywhere.
Bring questions. Etay will take them live.
What we’ll cover
- How agents failWhy agent mistakes look different from bugs, and why they’re harder to catch.
- Talked into itHow the content an agent reads can steer what it does next.
- What defenders can doPractical guardrails for teams putting agents to work today.
Etay Maor
Etay Maor is VP of Threat Intelligence at Cato Networks, an adjunct professor at Boston College Law School and a lecturer at MIT. He was previously an Executive Security Advisor at IBM, where he led security and fraud-fighting awareness and research.