Fact · Voice privacy and security
48percent
Open-source audio deepfake detectors lost 48% of their AUC, a measure of accuracy, on the 2024 in-the-wild benchmark compared with earlier benchmarks.
Checked against its source on October 17, 2026
In the source's own words
“AUC decreasing by 50% for video, 48% for audio, and 45% for image models compared to previous benchmarks”
Deepfake-Eval-2024: A Multi-Modal In-the-Wild Benchmark of Deepfakes Circulated in 2024, Nuria Alina Chandra et al., arXiv, 2025, Abstract.
Open the original- What it measures
- Deepfake-Eval-2024 collects deepfakes that circulated online in 2024, in 52 languages, and compares detectors on them with their scores on older academic datasets.
- What it does not show
- AUC is an aggregate score. The drop is reported for the group of open-source models, not for any one product.