The companies warning the world that AI needs to slow down are the same ones still selling access to it. Matthew de Mota, research director at the Canadian Shield Institute, untangles how much of that warning is genuine and how much doubles as a pitch for how powerful, and expensive, these tools really are.

Somewhere between the two extremes sits a real risk: tools capable of scanning an entire system for weaknesses can just as easily be pointed at breaking one instead. De Mota lays out what that looks like when the target shifts from a company's servers to a power grid or a water system.

Most people are using a version of these tools that barely resembles what's being tested behind closed doors. De Mota draws the actual line between the two, and explains why it matters more now than it has in previous years.

Topics: AI risk, cybersecurity, critical infrastructure, AI regulation, frontier technology

GUEST: Matthew de Mota

Originally aired on 2026-09-14