ML Security PapersBeyond Surface Artifacts: Capturing Shared Latent Forgery Knowledge Across Modalities
Jingtong Dou 1, Chuancheng Shi 1, Jian Wang 2, Fei Shen 3, Zhiyong Wang 1, Tat-Seng Chua 3
Published on arXiv
2604.07763
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
Achieves significant performance breakthroughs on unknown modalities by extracting universal forgery traces that transcend physical representations
MAF
Novel technique introduced
As generative artificial intelligence evolves, deepfake attacks have escalated from single-modality manipulations to complex, multimodal threats. Existing forensic techniques face a severe generalization bottleneck: by relying excessively on superficial, modality-specific artifacts, they neglect the shared latent forgery knowledge hidden beneath variable physical appearances. Consequently, these models suffer catastrophic performance degradation when confronted with unseen "dark modalities." To break this limitation, this paper introduces a paradigm shift that redefines multimodal forensics from conventional "feature fusion" to "modality generalization." We propose the first modality-agnostic forgery (MAF) detection framework. By explicitly decoupling modality-specific styles, MAF precisely extracts the essential, cross-modal latent forgery knowledge. Furthermore, we define two progressive dimensions to quantify model generalization: transferability toward semantically correlated modalities (Weak MAF), and robustness against completely isolated signals of "dark modality" (Strong MAF). To rigorously assess these generalization limits, we introduce the DeepModal-Bench benchmark, which integrates diverse multimodal forgery detection algorithms and adapts state-of-the-art generalized learning methods. This study not only empirically proves the existence of universal forgery traces but also achieves significant performance breakthroughs on unknown modalities via the MAF framework, offering a pioneering technical pathway for universal multimodal defense.
Key Contributions
- First modality-agnostic forgery (MAF) detection framework that decouples modality-specific artifacts to extract shared latent forgery knowledge
- Two progressive evaluation paradigms: Weak MAF (semantically correlated unseen modalities) and Strong MAF (completely isolated 'dark modalities')
- DeepModal-Bench benchmark integrating diverse multimodal forgery detection algorithms for systematic evaluation
🛡️ Threat Analysis
Core contribution is detecting AI-generated/manipulated content (deepfakes) across multiple modalities and verifying content authenticity. The paper proposes methods to identify synthetic content by capturing universal forgery traces left by generative models, which is directly about output integrity and AI-generated content detection.