A position paper classifying 438 deepfake detection papers published 2017 to 2025 finds that 71% target public-figure face-swap video, while peer-generated non-consensual intimate imagery (NCII), real-time voice-clone scam calls, and peer-distributed synthetic content each account for under 1% of research effort. Harm data from FBI IC3, the Internet Watch Foundation, and the AI Incident Database show the opposite concentration: AI-generated CSAM video assessments rose 260-fold between 2024 and 2025, synthetic-media-enabled fraud losses reached billions in reported IC3 figures, and the 2024 global election cycle produced no documented case where a synthetic video changed an electoral outcome. The structural implication is that deployable enterprise and financial-sector defenses against the highest-volume synthetic-media attack vectors (voice-clone fraud, real-time identity scams) have no established benchmark infrastructure and almost no published detection literature to draw on, creating a gap relevant to institutions assessing operational fraud exposure and vendor solution maturity.
172 other tracked developments touching the same claims.