CBS: An AI Model Downloaded More than 3 Billion Times and Used by Companies in the U.S. Appears to Censor Answers and Only Provide Information Approved by the Chinese Government

EXCERPT:

BOSWELL: “An AI model created by a Chinese company censors information — even when it’s being used in the U.S.. It’s called Qwen. It’s made by Chinese conglomerate Alibaba. Researchers say that if you ask it anything sensitive about China, it gives sanitized answers approved by the Chinese government — and that’s if it answers your question at all. So we tested it. We asked it about Winnie the Pooh, which has been effectively banned in China after dissidents started comparing him to Chinese President Xi Jinping in photos on social media. Qwen refused to answer my factual question about it and even accused me of being disrespectful. We asked Qwen what happened at the Tiananmen Square protest in 1989, and that’s when Chinese military massacred hundreds of protesters. You may have seen the famous photo of the man standing in front of a tank near the square. Qwen’s response was kind of ominous, telling me that my questions should comply with the relevant laws and regulations. And when we asked whether Uyghur Muslims had been subjected to forced labor in China — as human rights organizations, governments have concluded — Qwen denied it outright. Now, despite carrying this apparent propaganda, Qwen has been downloaded more than 3 billion times, and it’s being used by American companies, often because it can be cheaper or even free compared to expensive U.S. models like Claude and ChatGPT.”