aiqbench analysis
This research addresses the critical security challenge of detecting AI-generated videos by introducing a framework that relies on intrinsic forensic cues rather than spurious domain-specific biases. By employing counterfactual intervention and causal disentanglement, G2VD significantly improves cross-domain generalization while reducing training data requirements to just 10% compared to baselines. The approach offers a robust solution for maintaining detection integrity as generative models evolve beyond their original distribution.
G2VD: Generalizable AI-Generated Video Detection via Counterfactual Intervention and Causal Disentanglement
arXiv:2607.04607v2 Announce Type: replace-cross Abstract: Rapid advances in AI video generation pose increasing security risks and call for reliable detectors with strong cross-domain generalization. Although existing methods perform well under in-domain evaluation, their performance degrades substantially on unseen generators. A key reason is shortcut learning, where detectors rely on domain-specific bias rather than intrinsic forensic cues. To address this issue, we propose G2VD, a generalizable AI-generated video detection framework based on counterfactual intervention and causal disentanglement. First, G2VD introduces a counterfactual intervention pipeline (CFIPipeline) that constructs counterfactual samples through VAE-based reconstruction and subsequent frequency-domain and pixel-domain alignment, thereby weakening spurious correlations between domain-specific bias and authenticity labels. Building on this intervention, we further design a causal disentanglement classifier that combines two domain-anchored branches with complementary objectives and a constraint based on the Hilbert-Schmidt Independence Criterion (HSIC), encouraging the causal and non-causal representations to capture intrinsic forensic cues and domain-specific bias, respectively. Experiments across four public datasets demonstrate strong cross-domain performance and consistent gains over baseline methods. In the challenging GenVidBench setting, G2VD achieves over 90\% overall ACC, with improvements of 0.194 in F1 and 0.104 in AUC over comparable state-of-the-art methods, while using only 10\% of the available training data. Code is available at https://github.com/DMOSCAR-98/G2VD.
Abstract:Rapid advances in AI video generation pose increasing security risks and call for reliable detectors with strong cross-domain generalization. Although existing methods perform well under in-domain evaluation, their performance degrades substantially on unseen generators. A key reason is shortcut learning, where detectors rely on domain-specific bias rather than intrinsic forensic cues. To address this issue, we propose G2VD, a generalizable AI-generated video detection framework based on counterfactual intervention and causal disentanglement. First, G2VD introduces a counterfactual intervention pipeline (CFIPipeline) that constructs counterfactual samples through VAE-based reconstruction and subsequent frequency-domain and pixel-domain alignment, thereby weakening spurious correlations between domain-specific bias and authenticity labels. Building on this intervention, we further design a causal disentanglement classifier that combines two domain-anchored branches with complementary objectives and a constraint based on the Hilbert-Schmidt Independence Criterion (HSIC), encouraging the causal and non-causal representations to capture intrinsic forensic cues and domain-specific bias, respectively. Experiments across four public datasets demonstrate strong cross-domain performance and consistent gains over baseline methods. In the challenging GenVidBench setting, G2VD achieves over 90\% overall ACC, with improvements of 0.194 in F1 and 0.104 in AUC over comparable state-of-the-art methods, while using only 10\% of the available training data. Code is available at this https URL.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2607.04607 [cs.CV]
(or arXiv:2607.04607v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2607.04607 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 300 words
Related coverage
Game Design Creativity evals
LessWrong · Oct 12, 2026
Common Cause, Not Cross-Attention: Blocking Visual Shortcuts in Audio-Video Generation
arXiv · Sep 22, 2026
Links #5: 2026/07
LessWrong · Jul 31, 2026
Faithful Action-unit Causal Reasoning for Counterfactually Faithful Emotion Explanations
arXiv · Jun 16, 2026