Preprint finds labels on only 33% of high-risk deepfakes

A new preprint titled "Drowning in AI Slop" reports that only 33% of deepfakes sampled in systemic-risk contexts contained visible labels. The study, which has not yet undergone peer review, suggests a significant gap between voluntary industry commitments and actual compliance in high-stakes environments. Researchers analyzed content specifically categorized as posing systemic risks rather than general entertainment media. The findings indicate that current detection and transparency mechanisms fail to adequately tag the majority of potentially harmful synthetic media. Industry stakeholders have previously advocated for voluntary watermarking, but this data implies such measures remain inconsistently applied where consequences are highest. Policymakers considering mandatory disclosure laws may cite these preliminary findings as evidence of market failure. The authors caution that results are provisional pending formal academic validation. Viewers are advised to treat unlabeled viral content with skepticism regardless of source credibility.