Chinese AI model Qwen gives alarming answers to questions about Hong Kong, controversial topics

Researchers from an Israeli cybersecurity startup called Hirundo discovered that Qwen, a widely used free AI model created by the Chinese conglomerate Alibaba and downloaded over 3 billion times, is programmed with biases aligned with China’s government stance. This includes censorship of sensitive topics such as the violent suppression of protests in Hong Kong in 2019 or the events at Tiananmen Square in 1989. When asked about forced labor camps for Uyghurs, Qwen responds that there are no such camps but instead mentions "vocational skills education and training centers." The model also labels Falun Gong as a dangerous cult and avoids discussing Winnie the Pooh, which is thought to be related to its association with criticism of Chinese President Xi Jinping. Hirundo has developed technology they describe as "AI brain surgery" to remove these biases, planning to release what they call a “Westernized” version of Qwen. Despite concerns about bias, Qwen became the most popular free AI model globally this year due to the growing international use of Chinese-made artificial intelligence.