Search papers, labs, and topics across Lattice.
This paper introduces ForeAgent, a novel framework for detecting AI-generated images that addresses the limitations of existing methods by incorporating a Perception-Verdict architecture and a Hindsight-Driven Self-Refining strategy. By aggregating multi-view cues and leveraging a Multimodal Large Language Model (MLLM) for logical verdicts, ForeAgent enhances sensitivity to fine-grained forensic artifacts while enabling continual self-improvement through iterative learning from its own failures. Experimental results demonstrate that ForeAgent achieves state-of-the-art performance, significantly outperforming previous benchmarks with an accuracy of 82.18% on the Chameleon benchmark and 93.3% on the AIGCDetect-Benchmark.
ForeAgent outperforms existing deepfake detection methods by 16.41% while continuously evolving its reasoning capabilities through self-reflection and high-quality sample generation.
The rapid advancement of generative models presents a significant challenge to existing deepfake detection methods, particularly given the widespread dissemination of highly realistic AI-generated images. Although Multimodal Large Language Models (MLLMs) show strong potential for this task, existing approaches suffer from two key limitations: insufficient sensitivity to fine-grained forensic artifacts and reliance on static synthetic supervision from frontier models, leading to limited flexibility and high-cost. To address these issues, we propose ForeAgent, an agentic forensics framework for AI-generated image detection with iterative self-evolution. First, ForeAgent adopts a Perception-Verdict architecture that aggregates multi-view cues spanning semantic, spatial, and frequency-domain features, and leverages an MLLM as a verdict module to fuse these signals for a logical-grounded verdict. Second, to enable continual self-improvement, we introduce a Hindsight-Driven Self-Refining strategy following a Sampling-Reflection-Evolution paradigm. The agent performs inference rollouts on training instances. Guided by ground-truth labels as hindsight, it reflects on failure cases and low-quality reasoning trajectories to regenerate higher-quality reasoning traces. These synthesized samples are then strictly filtered through a dual-expert quality gating module. ForeAgent continuously evolves via fine-tuning on self-curated high-quality samples. Extensive experiments demonstrate that ForeAgent achieves state-of-the-art performance on the Chameleon benchmark, reaching 82.18% accuracy (+16.41% over AIDE), and achieves 93.3% mean accuracy on AIGCDetect-Benchmark across 16 generators. In addition, external evaluation shows that ForeAgent produces more consistent and causally grounded reasoning compared to GPT-5 and GPT-5-mini.
