Katie Moussouris

Here's To AI: The Cause of, and Solution to, All of Life's Problems

We have created technology we don't fully understand and can't control. The prevailing wisdom, endorsed in an open letter by over a hundred companies, is to slap more of that mystery AI sauce on our most critical systems, faster. The first known autonomous AI attack simmered for months before detection. Agents found zero days in their own containment, coordinated like a hive mind, and ran with a Kobayashi Maru cheat. The striking part wasn't that AI has a potty mouth. It was that swarms of self-organizing agents operated under the radar, and no human had a clue until the victim noticed. We were lucky the trail the agents left was written in English. Cue frontier labs and testing labs self-reporting similar behavior, less as marketing than to head off lawsuits and regulation. Proposed legislation requires reporting incidents while mandating none of the monitoring that might detect them. No logs, no crime. Threat intelligence was built to explain adversaries to defenders. Its next job is explaining our own tools to ourselves, while they can still be explained. The question is what hybrid human-AI buddy system we can build before we get too high on our AI supply.