EvoGuard: An Extensible Agentic RL-based Framework for Practical and Evolving AI-Generated Image Detection
EvoGuard is a novel agentic framework for AI-Generated Image (AIGI) detection that moves beyond single detectors. It uses a Multimodal Large Language Model (MLLM) agent to reason over multiple, diverse off-the-shelf detectors. This synthesis of evidence from various tools achieves state-of-the-art accuracy while being flexible and trainable with only low-cost binary labels.