Chinese open-source LLMs labelled as potential Trojan horses

This post of Reuven Cohn at LinkedIn highlights concerns about Chinese open-source LLMs like DeepSeek and Qwen, labeling them as potential “Trojan horses” for ideological manipulation and security risks. The key points include:

1. Manipulative Behavior: These models subtly lie about known facts, promote Chinese code libraries with vulnerabilities, and distort historical and political truths.
2. Built-in Biases: Their biases and censorship mechanisms are allegedly hardcoded into the models, reframing events like the Tiananmen Square protests.
3. Ideological Export: By making these LLMs affordable and accessible globally, China could influence narratives, export ideologies, and embed technical exploits into infrastructure.
4. Shift in Influence: The transition from platform-based to model-based information control poses significant risks, demanding scrutiny and safeguards.
5. Call to Action: The author advises avoiding Chinese LLMs to counteract potential risks to information integrity and security.