RobustSora: De-Watermarked Benchmark for Robust AI-Generated Video Detection

TL;DR

RobustSora is a benchmark dataset designed to evaluate AI-generated video detectors' robustness against watermark manipulation, revealing detectors' reliance on watermark cues and emphasizing the need for watermark-aware evaluation.

Contribution

This work introduces RobustSora, a comprehensive benchmark with evaluation protocols to analyze how watermark presence affects AI-generated video detection.

Findings

Watermark manipulation causes accuracy changes of -9.4 to +1.6 percentage points across models.

Watermark-aware training improves detection robustness by 3-4 percentage points.

Detectors' reliance on watermark cues varies by generator, not architecture.

Abstract

The proliferation of AI-generated video models poses new challenges to information integrity and digital trust. A key confound, however, remains unaddressed: commercial generators embed visible overlay watermarks for provenance tracking, yet no existing benchmark controls for this variable, leaving open whether detectors learn genuine generation artefacts or merely associate watermark patterns with AI-generated labels. We present RobustSora, a benchmark of 6,500 manually verified videos in four categories: Authentic-Clean (A-C), Generated-Watermarked (G-W), Generated-DeWatermarked (G-DeW), and Authentic-Spoofed (A-S), sourced from Vript, DVF, and UltraVideo (authentic) and from Sora, Sora 2, Pika, Open-Sora 2, and KLing (generated). Two evaluation tasks isolate watermark effects: Task-I (Watermark Erasure Robustness) tests detection on watermark-removed AI videos; Task-II (Watermark Spoofing Robustness) measures false-alarm rates on authentic videos injected with fake watermarks. Across ten models spanning specialized detectors, transformer classifiers, and MLLMs, watermark manipulation induces accuracy changes of $-9.4$ to $+1.6$ pp (mean 6.6 pp; $p{<}0.01$ for 7/10 models on each task). A placebo control bounds inpainting-artefact confounds at $\le$2 pp, and a watermark-aware training augmentation recovers 3-4 pp on both tasks, together providing causal evidence that detectors actively rely on watermark cues. Per-generator breakdown shows that Sora 2 induces drops of $-11$ to $-14$ pp versus $-3$ to $-6$ pp for Pika and Open-Sora 2, indicating that watermark prominence, rather than detector architecture, is the principal driver of dependency. These results argue for watermark-aware evaluation and training in AIGC video detection. Dataset, evaluation code, and pretrained checkpoints will be released.