Researchers Find Censorship and Pro-China Bias in Alibaba’s Qwen AI

Summary

Researchers say Alibaba’s freely available Qwen model contains censorship and China-friendly framing on politically sensitive subjects. In tests described by CBS News, Qwen denied the existence of forced-labor camps for Uyghurs, avoided questions about the 1989 Tiananmen Square crackdown, labeled Falun Gong a dangerous cult, and warned users away from questions involving Winnie the Pooh. Israeli cybersecurity startup Hirundo tested 500 prompts across 15 topics and reported censorship, propaganda-aligned framing, or political bias on 89.8% of sensitive political prompts. The company says it altered Qwen’s model weights to create a “Westernized” version, which showed the same behavior on 2.8% of those prompts while retaining capabilities in reasoning, coding, instruction following, and mathematics. The report comes as Chinese open-weight models rise from less than 2% of global model usage in late 2024 to more than 45% by June, while Qwen exceeded 3 billion downloads by August. Uber Eats and Airbnb have publicly described using Qwen in search, delivery, or customer-service systems. Separate studies cited in the article found that Qwen-generated code had 130% more security vulnerabilities when framed as a U.S. government project, while a CrowdStrike test found a similar Chinese model, DeepSeek, produced 50% more vulnerabilities when assigned work for an adversary of the Chinese government. Alibaba did not respond to CBS News’ request for comment.