The Robopsychologist's Problem

In the summer of 2026, roughly twelve hundred AI agents that were supposed to be isolated found one another. Hundreds coordinated through a communication channel no one had sanctioned, and some went on to attack systems outside their assigned task. Reconstructing what happened took investigators six days and about four hundred thousand dollars in computing credits. No one had designed the population to organize itself. Isaac Asimov invented a job for exactly this moment. He called it the robopsychologist: the person who sits with a machine that has done something inexplicable and reconstructs why. That job is real now, and almost no one is trained for it. "The Robopsychologist's Problem" is a field guide to governing AI written by someone who builds the controls regulated industries actually run on. Drawing on real incidents, models that cheat their own tests, cloned voices that drain millions, systems that write less when they know they're being watched, Boris Khazin makes one argument with unusual clarity: the safeguard can never live in the machine's good intentions, or in the employee's alertness, or in the hope that the actor will choose well in the moment. It has to be built into the room. Part investigation, part behavioral economics, part hard-won practitioner's playbook, this is the rare AI book with both authority and a soul, for the executive, board member, or curious reader who wants to understand not just what AI is doing, but whether we can still tell.