Deceptive synthetic media threaten the credibility of information at critical moments. AI detection tools can help mitigate this risk—but only if they are effective, accessible, and usable in real-world, high-stakes contexts.
The DeFake team are happy to have contributed to this effort that led to WITNESS’ TRIED: Truly Innovative and Effective AI Detection Benchmark. This document introduces a new sociotechnical framework to evaluate AI detection tools not just on accuracy, but on their usability and effectiveness in diverse global situations. Grounded in real-world cases of deceptive AI and shaped by global consultations, the TRIED Benchmark outlines clear, actionable steps to guide the design, deployment, and standards of detection tools that are resilient, innovative, and globally relevant.
📄 Read the full report on arXiv: https://arxiv.org/abs/2504.21489