ML Security PapersRAIDX: A Retrieval-Augmented Generation and GRPO Reinforcement Learning Framework for Explainable Deepfake Detection
Tianxiao Li 1, Zhenglin Huang 1, Haiquan Wen 1, Yiwei He 1, Shuchang Lyu 2, Baoyuan Wu 3, Guangliang Cheng 1
Published on arXiv
2508.04524
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
RAIDX achieves state-of-the-art deepfake detection performance while autonomously generating interpretable textual rationales and saliency maps, eliminating the need for manual annotation
RAIDX
Novel technique introduced
The rapid advancement of AI-generation models has enabled the creation of hyperrealistic imagery, posing ethical risks through widespread misinformation. Current deepfake detection methods, categorized as face specific detectors or general AI-generated detectors, lack transparency by framing detection as a classification task without explaining decisions. While several LLM-based approaches offer explainability, they suffer from coarse-grained analyses and dependency on labor-intensive annotations. This paper introduces RAIDX (Retrieval-Augmented Image Deepfake Detection and Explainability), a novel deepfake detection framework integrating Retrieval-Augmented Generation (RAG) and Group Relative Policy Optimization (GRPO) to enhance detection accuracy and decision explainability. Specifically, RAIDX leverages RAG to incorporate external knowledge for improved detection accuracy and employs GRPO to autonomously generate fine-grained textual explanations and saliency maps, eliminating the need for extensive manual annotations. Experiments on multiple benchmarks demonstrate RAIDX's effectiveness in identifying real or fake, and providing interpretable rationales in both textual descriptions and saliency maps, achieving state-of-the-art detection performance while advancing transparency in deepfake identification. RAIDX represents the first unified framework to synergize RAG and GRPO, addressing critical gaps in accuracy and explainability. Our code and models will be publicly available.
Key Contributions
- First unified RAG+GRPO framework for explainable deepfake detection, integrating external knowledge retrieval with reinforcement learning-driven rationale generation
- Autonomous generation of fine-grained textual explanations and saliency maps without labor-intensive manual annotations
- State-of-the-art detection accuracy across multiple deepfake benchmarks with improved transparency
🛡️ Threat Analysis
Deepfake detection is AI-generated content detection — an explicit ML09 concern. RAIDX proposes a genuinely novel detection architecture (RAG + GRPO) for identifying real vs. AI-generated images, advancing both detection accuracy and output-level explainability via saliency maps and textual descriptions. This is not a mere domain application of existing methods; the framework itself is the primary contribution.