VeriTaS: A Living Benchmark for Multimodal Fact-Checking

Misinformation is growing fast, but today’s AI fact-checkers are often graded on outdated, static tests that leak into model pretraining—making scores look better than real-world performance.

VeriTaS is a living benchmark for automated fact-checking across text, images, and video. It stays fresh and robust, so evaluations actually mean something.

• 24,000 real claims from 108 fact-checking orgs, spanning 54 languages
• Multimodal: textual and audiovisual content
• Quarterly updates via a 7‑stage pipeline that normalizes claims, fetches original media, and maps expert verdicts to a clear, disentangled scoring scheme with explanations
• Human evaluations show automated labels closely match expert judgments
• Open data and code; designed to resist pretraining leakage

For researchers, journalists, and industry: use VeriTaS to test AFC systems against the world as it is—now and tomorrow. More: https://arxiv.org/abs/2601.08611

Paper: https://arxiv.org/abs/2601.08611v1

Register: https://www.AiFeta.com

#AI #FactChecking #Misinformation #Benchmark #Multimodal #NLP #ComputerVision #LLMs #OpenScience