The consensus threat model for AI deepfakes was built around a specific scenario: synthetic media deployed in coordinated campaigns to manipulate voter behavior at scale. That model is not wrong — the Reuters coverage of the 2026 midterms confirms it [2] — but it trained attention and resources on a surface that turned out to be one of several, not the primary one. The AI-generated fake doctors endorsing supplements , the voice-clone phone fraud, the synthetic romantic personas — these are not edge cases or secondary concerns. They are the…