ML Security PapersRA-Det: Towards Universal Detection of AI-Generated Images via Robustness Asymmetry
Xinchang Wang 1, Yunhao Chen 2, Yuechen Zhang 1, Congcong Bian 1, Zihao Guo 1, Xingjun Ma 2, Hui Li 1
Published on arXiv
2603.01544
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
RA-Det improves average AI-generated image detection performance by 7.81% over more than 10 strong competing detectors across 14 diverse generative models, with no generator-specific fingerprints required.
RA-Det (Robustness Asymmetry Detection)
Novel technique introduced
Recent image generators produce photo-realistic content that undermines the reliability of downstream recognition systems. As visual appearance cues become less pronounced, appearance-driven detectors that rely on forensic cues or high-level representations lose stability. This motivates a shift from appearance to behavior, focusing on how images respond to controlled perturbations rather than how they look. In this work, we identify a simple and universal behavioral signal. Natural images preserve stable semantic representations under small, structured perturbations, whereas generated images exhibit markedly larger feature drift. We refer to this phenomenon as robustness asymmetry and provide a theoretical analysis that establishes a lower bound connecting this asymmetry to memorization tendencies in generative models, explaining its prevalence across architectures. Building on this insight, we introduce Robustness Asymmetry Detection (RA-Det), a behavior-driven detection framework that converts robustness asymmetry into a reliable decision signal. Evaluated across 14 diverse generative models and against more than 10 strong detectors, RA-Det achieves superior performance, improving the average performance by 7.81 percent. The method is data- and model-agnostic, requires no generator fingerprints, and transfers across unseen generators. Together, these results indicate that robustness asymmetry is a stable, general cue for synthetic-image detection and that carefully designed probing can turn this cue into a practical, universal detector. The source code is publicly available at Github.
Key Contributions
- Identifies 'robustness asymmetry' as a universal behavioral signal: natural images maintain stable semantic embeddings under small structured perturbations while AI-generated images exhibit markedly larger feature drift
- Provides theoretical analysis establishing a lower bound connecting robustness asymmetry to memorization tendencies in generative models, explaining its generality across architectures
- Introduces RA-Det, a data- and model-agnostic detection framework requiring no generator fingerprints that achieves +7.81% average improvement over 10+ detectors across 14 generative models
🛡️ Threat Analysis
Proposes a novel detection framework for AI-generated (synthetic) images — a direct contribution to output integrity and content authenticity. The paper introduces a new behavioral forensic signal (robustness asymmetry) and a detector that outperforms 10+ existing methods across 14 generative models.