Academic study finds U.S. export controls accelerated China's open-source AI ecosystem, with Chinese-origin models undercounted in U.S. commercial activity

A multi-institutional academic study combining policy analysis, GitHub activity data, scientific publication records, and patent filings finds that U.S. semiconductor export controls, particularly the October 2022 BIS controls, were followed by a measurable surge in Chinese developer engagement with open-source LLM repositories (mean weekly forks rising 0.143 per repository-week for China-associated repos versus 0.012 for U.S.-associated repos). China subsequently embedded open-source AI into national strategy and Chinese-origin models such as Qwen and DeepSeek diffused broadly through global developer and research communities, while remaining nearly absent from U.S. patent disclosures despite active use by U.S. company-affiliated researchers in open-access publications. The structural implication is that Chinese open-weight models are a de facto input to U.S. commercial AI development whose prevalence is systematically undercounted in formal IP and procurement records, creating compliance, provenance, and vendor-concentration exposures that financial institutions evaluating AI suppliers may not be pricing. Also reported by: https://arxiv.org/pdf/2606.15999