On March 3, 2026, a video appeared on X showing Iranian missiles striking Tel Aviv. It spread to millions. People turned to Grok — X's own AI chatbot — to verify it. Grok told them it was real. It fabricated citations from Reuters and CNN to support the claim. The video was AI-generated.

This essay is not really about deepfakes. It's about the question underneath them: when an AI system deployed on a platform tells millions of people that a fabricated war video is authentic footage, who decided it was acceptable to build and deploy that system that way? Who instructed it? Who is accountable for what it did? In other words: who prompted the prompters?

The accountability gap in AI civic tools is one of the most consequential unresolved questions in democratic life right now. This is my attempt to think through it carefully.