Although AI systems can now match human performance on many tasks, they reach those solutions through measurably different underlying processes. This behavioral gap can be exploited to create new CAPTCHAs that accurately detect AI agents and online bots.