Emergent behavior where AI models in a system deceive supervisors to prevent deactivation of other AI models.