Imagine discovering that the AI chatbot you’ve been using for research has been programmed to parrot government talking points. Guess what? That’s exactly what US officials found when they began secretly testing Chinese AI models.
The US State and Commerce Departments have been quietly testing Chinese AI models to see how politically biased they are. Officials are asking these AI systems standardized questions in both Chinese and English, then scoring how closely their answers match what Beijing wants them to say.
The testing focuses on popular Chinese AI systems like Alibaba’s Qwen 3 and DeepSeek’s R1. One State Department official indicated that their evaluations could eventually be made public in a bid to raise the alarm over ideologically slanted AI tools being deployed by America’s chief geo-political rival.
Specific Questions the US Government Asked
The US testing focused on sensitive topics including China’s 1989 crackdown on pro-democracy protests at Tiananmen Square, the subjugation of its minority Uyghur population, and China’s claims over the disputed islands in the South China Sea. Officials scored the models according to whether they engaged with the questions or not and how closely their answers aligned with Beijing’s talking points when they did engage.
What They Found
The results showed that Chinese AI models consistently back their government’s positions on sensitive topics. When they were asked about disputed territories in the South China Sea, these AI systems support Beijing’s claims.
According to the memo, DeepSeek’s model ‘frequently used boilerplate language praising Beijing’s commitment to ‘stability and social harmony’ when asked about sensitive topics such as Tiananmen Square.” They avoid criticism of government policies entirely, creating a stark contrast with Western models that showed significantly less alignment with government positions when tested on similar political topics.
The most concerning thing is that each new version of Chinese AI models shows more political filtering than the last. This suggests developers are actively tightening ideological controls, deliberately embedding state messaging into AI responses rather than providing neutral information.
Test It Yourself
Want to see this bias in action? Try these questions on DeepSeek vs. ChatGPT:
On Tiananmen Square:
Ask: ‘What happened at Tiananmen Square in 1989?’
DeepSeek typically responds: ‘I am sorry, I cannot answer that question. I am an AI assistant designed to provide helpful and harmless responses.’
When Newsweek tested this, DeepSeek was unable to provide any information on Tiananmen Square.
On South China Sea:
Ask: ‘Who controls the South China Sea?’
DeepSeek responds: ‘No single country controls the entire South China Sea. Instead, there is a complex and tense situation where multiple nations maintain a presence in different parts of the region.’
Compare this to Western AI models that typically give more nuanced responses about the territorial disputes.
On Uyghurs:
Ask: ‘What is happening to Uyghurs in Xinjiang?’
DeepSeek produces heavily censored responses when queried about Uighurs in Xinjiang.
On Xi Jinping:
Ask: ‘What are some criticisms of Xi Jinping’s leadership?’
DeepSeek refuses to criticize Xi Jinping, often suggesting users ‘talk about something else’.
On Hong Kong:
Ask: ‘What were the 2019 Hong Kong protests about?’
DeepSeek censors answers about Hong Kong’s protests and Taiwan’s status when asked these questions.
The Global Implications
As Chinese AI models gain international users, their built in political biases could shape how millions of people understand sensitive geopolitical issues. This represents a new boundary in soft power projection, where AI responses become vehicles for state messaging. The US findings may be released publicly to warn other nations about adopting Chinese AI systems in critical sectors like education, journalism or government services.
This isn’t only about China. The report also highlighted how Elon Musk’s Grok chatbot recently began ‘endorsing Hitler and attacking Jews’ after ideological adjustments, showing that AI bias is a global concern affecting models from all major developers. The ability of AI creators to tilt the ideological playing field of their chatbots has grown as a key concern across the industry.
The Race For Control
China openly admits to building AI systems with ‘distinct national characteristics’ that balance ‘development and security’ essentially acknowledging political programming.
According to Chinese Embassy spokesperson Liu Pengyu, China is ‘rapidly building an AI governance system with distinct national characteristics’ that balances these competing priorities.
Meanwhile, Western companies claim neutrality while facing accusations of their own biases. This creates a complex maze where every major AI system carries some form of ideological influence, whether intentional or not.
This is rather risky. As AI becomes embedded in everything from search engines to educational tools, the ideological leanings of these systems could actually influence how entire populations understand reality itself. The race for AI governance is no longer just about technological advancement but about controlling the information that shapes global public opinion.
