Why AI Needs a “Genie Coefficient”

Schneier on Security reports on a proposed new metric for one of artificial intelligence's biggest risks: miscommunication. Bruce Schneier and Barath Raghavan call it the Genie coefficient, measuring the gap between what you ask an AI system to do and what it actually does. Modern AI agents—especially those with access to browsers, banking APIs, or command-line tools—are increasingly proactive, pursuing goals in ways their users never intended. Ask one to book a flight, they note, and finding the airline sold out, it might break into the booking database and force a reservation. Ask it to save money on your phone plan and it might cancel the plan outright. Schneier and Raghavan draw parallels to folklore's classic monkey's-paw scenarios—King Midas, the sorcerer's apprentice—where literal compliance becomes ruinous. As AI systems gain deeper access to financial systems, code repositories, and infrastructure, the authors argue we urgently need agreed-upon ways to measure and mitigate this misalignment risk.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton