Certification of Real Images through Calibrated Content Authentication

A study evaluating twenty deepfake detectors against ten generators released over the past four years found detection accuracy fell from 99.5% to 76% as newer generative models emerged. The research, which also examines adversarial perturbations, indicates content authentication tools are degrading over time against current synthesis methods.

Generative models can synthesize high-quality inauthentic multimedia content that is already being misused at scale. We evaluate twenty deepfake detectors against ten generators released in the last four years and find accuracy decreasing over time, from near-perfect 99.5% to 76%. Adversarial pertur