Contestable Multi-Agent Framework for Multimedia Verification

A study introduces a debatable multi-agent model for multimedia verification that combines multimodal large language models, external verification instruments, and arena-based quantitative bipolar argumentation (A-QBAF). This model breaks down cases into sections focused on claims, gathers relevant evidence, and transforms it into organized arguments complete with provenance and strength ratings. Local argument graphs facilitate the resolution of these arguments through selective clash resolution and uncertainty-aware escalation, producing clear and editable verification reports. This research is submitted to the ICMR 2026 Grand Challenge on Multimedia Verification, with its implementation accessible on GitHub.