Study Measures AI Washing in Chinese Capital Markets Using Multimodal Analysis

A study released on arXiv investigates the phenomenon of AI washing within the capital markets of China, where firms inflate their artificial intelligence abilities. The research covers China's A-share market from Q1 2018 to Q2 2025. Utilizing Qwen-VL, the researchers devised a multimodal AI Washing Risk Score to assess the alignment between text and images in annual reports and roadshow presentations. Additionally, they formulated a Material Real-Investment Matching Index, which considers patent quality, AI intangible asset capitalization, and compensation for technical staff through principal component analysis. The study reveals four significant insights: the AI Washing Risk Score fails to forecast future material investment matching; financially constrained companies exhibit a greater disparity between claims and actual investment; genuine AI investment results in superior patents; and hollow AI claims hinder industry innovation. The research highlights how information asymmetry and technological opacity facilitate the exaggeration of AI capabilities over authentic R&D, particularly as AI and generative large language models propel industrial advancement, drawing more attention from capital markets towards AI-focused firms.