Undoing the machine - when AI errors become a cyber risk

Artificial Intelligence (AI) agents are moving from pilot projects into production across almost every industry. They're automating tasks, integrating with enterprise systems, and in many cases making operational decisions without human oversight. That acceleration promises huge efficiency gains — but it also introduces a new category of risk that traditional cybersecurity strategies weren't designed to handle: "non-human error."

Richard Cassidy, EMEA CTO at Rubrik, has been speaking with Chief Information Security Officers (CISOs) across industries about this emerging threat. Speaking to him this week, he draws a sharp distinction between the errors of human operators and the mistakes of autonomous software agents:

What we're describing here is a new class of operational failure. Mistakes introduced not by distracted employees or mis-typed configurations, but by AI agents acting at speeds that we just can't fathom anymore.

The reality of non-human error

Cassidy shared examples that bring this risk ...