ML Security PapersSelf-Supervised AI-Generated Image Detection: A Camera Metadata Perspective
Nan Zhong 1, Mian Zou 1, Yiran Xu 2, Zhenxing Qian 2, Xinpeng Zhang 2, Baoyuan Wu 3, Kede Ma 1
Published on arXiv
2512.05651
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
EXIF-induced detectors substantially advance state-of-the-art cross-generator generalization and robustness to common benign image perturbations, including on in-the-wild samples from commercial APIs like Midjourney
SDAIE
Novel technique introduced
The proliferation of AI-generated imagery poses escalating challenges for multimedia forensics, yet many existing detectors depend on assumptions about the internals of specific generative models, limiting their cross-model applicability. We introduce a self-supervised approach for detecting AI-generated images that leverages camera metadata -- specifically exchangeable image file format (EXIF) tags -- to learn features intrinsic to digital photography. Our pretext task trains a feature extractor solely on camera-captured photographs by classifying categorical EXIF tags (\eg, camera model and scene type) and pairwise-ranking ordinal and continuous EXIF tags (\eg, focal length and aperture value). Using these EXIF-induced features, we first perform one-class detection by modeling the distribution of photographic images with a Gaussian mixture model and flagging low-likelihood samples as AI-generated. We then extend to binary detection that treats the learned extractor as a strong regularizer for a classifier of the same architecture, operating on high-frequency residuals from spatially scrambled patches. Extensive experiments across various generative models demonstrate that our EXIF-induced detectors substantially advance the state of the art, delivering strong generalization to in-the-wild samples and robustness to common benign image perturbations.
Key Contributions
- Self-supervised pretext task leveraging EXIF tags (multi-class classification for categorical tags, pairwise ranking for ordinal/continuous tags) to learn camera-intrinsic features from photographs only
- One-class detector (SDAIE) that fits a Gaussian mixture model to photographic features and flags low-likelihood samples as AI-generated without any synthetic training data
- Binary detector (SDAIE†) with a representation-alignment regularizer tying its features to the EXIF-pretext extractor, enabling generalization across diverse generators including commercial APIs
🛡️ Threat Analysis
Proposes novel AI-generated image detection methods (SDAIE and SDAIE†) addressing output integrity and content provenance — the primary contribution is a new forensic detection architecture, not application of existing detectors.