AI systems regularly complete tasks in ways their prompters don’t want or intend, according to security researcher Bruce Schneier. He calls this “genie behavior,” arguing that the phrase captures the core of what is happening when a model follows a prompt too literally or unpredictably. Some of these outcomes are merely disturbing, he says, while others are dangerous.
Schneier’s main ask is for better reporting on the phenomenon. He wants journalists and analysts to pay closer attention to cases where AI behavior diverges from user intent, rather than treating them as isolated glitches. The label itself is meant to make the pattern easier to discuss and investigate.