Governing the Invisible: Cognitive Externalities of AI and the Limits of Current Accountability Frameworks | Synapse