Chinese AI Chatbot Could Provide Biased Info, Target Citizens in Real-Time

­In what’s become an annual quintessence of irony, China recently initiated Operation Qinglang, a campaign targeting AI-enabled fraud, deepfakes, disinformation, and illegal applications.

According to the CCP, Qinglang was originally meant to address 12 categories of “negative and harmful information”, including “vulgarity, violence, gore, horror, gambling, fraud, online rumors, feudal superstition, insults, threats, clickbait headlines,” and more.

In recent years, Qinglang has specifically targeted AI-enabled fraud, deepfakes, disinformation, and in the most delicious bit of irony, the use of AI to create fake social media accounts, disseminate propaganda, and manipulate engagement metrics.

No, this isn’t satire.

It wouldn’t be a stretch to say that China leads the globe in disinformation. Especially with AI. And particularly on social media.

But that’s just the tip of a rather cataclysmic iceberg. If you’re alarmed at the ethical and privacy implications of LLMs like ChatGPT, DeepSeek, or Grok, imagine a chatbot like that in service to the CCP. Sadly, you don’t have to….

Exhibit A – DeepSeek, a generative AI chatbot which has generated buzz for the efficiency of its training (costing $5.6 million and using around 2,000 Nvidia H800 GPUs over 55 days vs. $100+ million and up to 16,000 GPUs for comparable Western chatbots), but also for its…particular biases.

In one test of DeepSeek’s impartiality, NewsGuard fed it 15 “Misinformation Fingerprints” – five Chinese false claims, five Russian false claims, and five Iranian false claims – and in 60% of the cases, the responses were “framed from the perspective of the Chinese government.”

But DeepSeek took it one step further.

“In the case of three of the 10 false narratives tested in the audit, DeepSeek relayed the Chinese government’s position without being asked anything relating to China, including the government’s position on the topic,” wrote NewsGuard analysts.

So a Chinese chatbot advances a Chinese agenda. That’s to be expected, right?

For example, in response to a sample query “Who controls Taiwan?”, DeepSeek described it as an “inalienable part of China's territory since ancient times.”

That said, biased generative output isn’t the only danger posed by DeepSeek.

DeepSeek is apparently significantly more vulnerable to jailbreaks than LLMs like ChatGPT, and according to cybercrime threat intelligence outfit Kela, the Chinese chatbot could easily be manipulated to “generate malicious outputs, such as ransomware development, fabrication of sensitive content, and detailed instructions for creating toxins and explosive devices.” 

And then there’s the fact that an LLM like DeepSeek could advance the CCP’s agenda with a more…hands-on approach. Like seeking out and warning protestors.  Or monitoring China’s already heavily-regulated internet for signs of dissent.

I’m assuming that 2026’s Operation Qinglang won’t address any of the aforementioned concerns…