Multi-Agentic System Leveraging Open-Source LLMs to Mitigate Disinformation Threats

Researchers have proposed a multi-agent artificial intelligence system built on open-source large language models to automate the detection of disinformation, arguing that manual fact-checking cannot keep pace with the scale of online falsehoods. The system, detailed in a paper submitted to arXiv on June 29, 2026, is designed to emulate the decision-making processes of human annotators by incorporating a consensus mechanism alongside diversity in cognition and knowledge [1][2]. The authors report that this approach achieved superior results compared to individual large language models, including GPT-4 and GPT-3.5, when tested on disinformation detection tasks [1][2]. The framework relies on open models such as LLaMA, Kimi, Qwen, DeepSeek, and LLaMA-Nemotron, a design choice the researchers said ensures greater transparency [1][2]. Evaluations were conducted across datasets in English, Polish, Slovak, and Bulgarian, spanning high-resource to low-resource languages [1][2]. The threat the system targets has been identified as a top-tier global risk. The World Economic Forum in January 2024 classified misinformation and disinformation as the most severe short-term global risks, warning they can widen societal and political divides [6]. The academic field of AI safety, which gained significant momentum in 2023 alongside rapid advances in generative AI, focuses on preventing harmful consequences from such systems, though researchers have expressed concern that safety measures are not keeping pace with development [3]. The use of open-weight models like DeepSeek has drawn industry attention since the Chinese company launched its R1 model in January 2025. DeepSeek reported training costs of $6 million for its V3 model, a fraction of the estimated $100 million cost for OpenAI's GPT-4, using weaker AI chips adapted to trade restrictions [8]. The company's chatbot briefly surpassed ChatGPT as the most downloaded freeware app on the U.S. iOS App Store [9]. OpenAI, which released ChatGPT in November 2022 and catalyzed the current AI boom, saw roughly half of its then-employed AI safety researchers depart throughout 2024, citing a deprioritization of safety goals [10].