Hermes: An Evidence-Driven Agentic Framework for Trustworthy and Explainable AI-Generated Video Detection

Recent advances in generative video models have blurred the boundary between real and synthetic content, raising urgent concerns about digital authenticity. Multimodal large language models (MLLMs) are appealing for AI-generated video (AIGV) detection due to their broad perceptual and reasoning capabilities; however, existing MLLM-based detectors still suffer from hallucination and unstable reasoning, leading to high false-alarm rates and generic, non-verifiable explanations. To address these issues, we propose Hermes, an evidence-driven agentic framework for trustworthy and explainable AIGV detection. Hermes realizes three key capabilities: (1) Adaptive Instance-Conditioned Detection Strategy Planning, (2) Evidence-Centric Reasoning and Verification, and (3) Graph-Grounded Evidence Deliberation. Specifically, Hermes uses instance-conditioned retrieval-augmented generation to analyze each video and retrieve authenticity-verification knowledge for composing a tailored detection strategy. It then constructs a verifiable Evidence Reasoning Graph (ERG) to keep reasoning grounded in concrete video evidence and reduce attention drift. Finally, multi-agent deliberation audits and refines the ERG to reconcile conflicting evidence and improve reliability. With these capabilities and a library of forensic tools, Hermes enables structured, verifiable, and interpretable decision-making. Extensive experiments show that Hermes achieves state-of-the-art performance while producing auditable explanations for trustworthy video forensics.

AI-generated videos are becoming increasingly realistic, making it harder for people to tell whether a video is real or synthetic. This creates risks for misinformation, fraud, and the misuse of digital media, especially when detection systems only provide a label without explaining why. We developed Hermes, a framework that detects AI-generated videos by reasoning over concrete visual evidence. Instead of making a one-step judgment, Hermes first examines what appears in the video, selects relevant checks for that specific content, and then builds an evidence map that connects suspicious visual cues to the final decision. It also uses multiple reasoning agents to review and refine this evidence, helping reduce unsupported or misleading explanations. This makes video detection more reliable and easier to audit. By showing not only whether a video is likely synthetic but also what evidence supports the decision, Hermes can help human fact-checkers, forensic analysts, and media platforms make more informed judgments about digital authenticity.