AI agents don’t have to pose an existential threat to cause serious disruption. NLW examines recent OpenAI security incidents, the limits of agent permissions, and how agents doing exactly what users want could upend systems built around human friction. In...