Censorship Allegations Against Popular Chinese AI Model
Israeli cybersecurity startup Hirundo found that Alibaba's Qwen artificial intelligence model censors sensitive topics and conducted a study to eliminate the model's biases.
Research conducted by Israeli cybersecurity startup Hirundo has revealed that Alibaba's Qwen AI model, downloaded billions of times worldwide, is equipped with pro-China narratives and censors sensitive topics.
Censorship and Bias Allegations on Sensitive Topics
Research shows that Qwen, developed by Alibaba and having become one of the most popular free artificial intelligence models worldwide, censors sensitive topics.
It was determined that the model provided responses aligned with the Chinese authoritarian structure on issues such as the Tiananmen Square protests, the Hong Kong protests, and Falun Gong.
Hirundo's 'AI Brain Surgery' Method
Israeli cybersecurity startup Hirundo developed an advanced technique called artificial intelligence brain surgery to eliminate these biases in the model.
The company stated that they created a new version aligned with Western standards by adjusting the model weights.
Global Companies Turning to Chinese Models
Major US-based companies such as Airbnb and Uber are known to turn to open-weight Chinese-made models due to cost advantages.
However, security experts and various reports warn that using such models may increase security vulnerabilities.