Proactive Defense Benchmark against Deepfake Generation

Joonhyuk Baek, Wonjune Seo, Jae-Yun Kim, Saerom Park, Hoki Kim
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:5144-5178, 2026.

Abstract

Despite the proliferation of proactive defenses against deepfakes, the lack of a unified evaluation protocol precludes fair comparison and masks critical vulnerabilities. To bridge this gap, we present the first comprehensive benchmark that systematically assesses disruption, robustness, and transferability encompassing pixel, perceptual, and identity metrics. Our extensive analysis reveals that fidelity and identity metrics capture orthogonal performance axes, often leading to conflicting interpretations when relied upon individually. Furthermore, we identify a fundamental trade-off where peak white-box performance signals overfitting, and we introduce a calibrated evaluation to correct generator-induced identity bias. By exposing these blind spots, we establish a rigorous standard to guide the development of genuinely generalizable protections. Project page is available at: https://proactivedefensebenchmark.github.io/

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-baek26e, title = {Proactive Defense Benchmark against Deepfake Generation}, author = {Baek, Joonhyuk and Seo, Wonjune and Kim, Jae-Yun and Park, Saerom and Kim, Hoki}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {5144--5178}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/baek26e/baek26e.pdf}, url = {https://proceedings.mlr.press/v306/baek26e.html}, abstract = {Despite the proliferation of proactive defenses against deepfakes, the lack of a unified evaluation protocol precludes fair comparison and masks critical vulnerabilities. To bridge this gap, we present the first comprehensive benchmark that systematically assesses disruption, robustness, and transferability encompassing pixel, perceptual, and identity metrics. Our extensive analysis reveals that fidelity and identity metrics capture orthogonal performance axes, often leading to conflicting interpretations when relied upon individually. Furthermore, we identify a fundamental trade-off where peak white-box performance signals overfitting, and we introduce a calibrated evaluation to correct generator-induced identity bias. By exposing these blind spots, we establish a rigorous standard to guide the development of genuinely generalizable protections. Project page is available at: https://proactivedefensebenchmark.github.io/} }
Endnote
%0 Conference Paper %T Proactive Defense Benchmark against Deepfake Generation %A Joonhyuk Baek %A Wonjune Seo %A Jae-Yun Kim %A Saerom Park %A Hoki Kim %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-baek26e %I PMLR %P 5144--5178 %U https://proceedings.mlr.press/v306/baek26e.html %V 306 %X Despite the proliferation of proactive defenses against deepfakes, the lack of a unified evaluation protocol precludes fair comparison and masks critical vulnerabilities. To bridge this gap, we present the first comprehensive benchmark that systematically assesses disruption, robustness, and transferability encompassing pixel, perceptual, and identity metrics. Our extensive analysis reveals that fidelity and identity metrics capture orthogonal performance axes, often leading to conflicting interpretations when relied upon individually. Furthermore, we identify a fundamental trade-off where peak white-box performance signals overfitting, and we introduce a calibrated evaluation to correct generator-induced identity bias. By exposing these blind spots, we establish a rigorous standard to guide the development of genuinely generalizable protections. Project page is available at: https://proactivedefensebenchmark.github.io/
APA
Baek, J., Seo, W., Kim, J., Park, S. & Kim, H.. (2026). Proactive Defense Benchmark against Deepfake Generation. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:5144-5178 Available from https://proceedings.mlr.press/v306/baek26e.html.

Related Material