China’s Propaganda Tool Disguised as AI Innovation

Until recently, Liang Wenfeng was a name virtually unknown. His meteoric rise from the world of hedge funds to one of the most talked-about names in artificial intelligence has dramatically captured the attention of the global tech elite. As the founder of DeepSeek, China’s latest and most ambitious large language model (LLM), Liang has positioned himself at the forefront of Beijing’s push for AI self-sufficiency. DeepSeek is being hailed as China’s answer to OpenAI’s GPT-4, a model that could rival, or even surpass, its Western counterparts. Yet beneath the technological prowess lies the troubling reality that DeepSeek is not just an AI breakthrough but an extension of the Chinese Communist Party’s (CCP) propaganda machine.

In the late 2010s, Liang co-founded a hedge fund that leveraged artificial-intelligence models to deliver impressive returns. His firm attracted billions of dollars, marking him as a shrewd investor and strategist in China’s fast-paced financial world. Despite the fund’s success, Liang was not immune to the shifting tides of Chinese regulation. He made a bold move by investing heavily in A.I., particularly in A.I. chips, and assembling a team with a singular mission to build China’s answer to Silicon Valley’s OpenAI.

Liang’s journey also mirrors the seismic shifts in China’s technology landscape over the past decade. His 2010 thesis at Zhejiang University focused on a subject that would later dominate the A.I. discourse in China: improving tracking algorithms for surveillance cameras. That early work would foreshadow a career increasingly centered around artificial intelligence, but it was his decision to seize the moment in 2023 in turning his focus to A.I. chips and generative models that may yet define him as one of China’s foremost tech pioneers.

From its inception, DeepSeek was designed to align with the CCP’s strict information controls. Unlike Western AI models, which, despite their own moderation policies, still allow room for open discourse, DeepSeek operates under Beijing’s censorship regime. It refuses to acknowledge politically sensitive events such as the Tiananmen Square massacre, parrots the state’s position on Taiwan and Tibet as inseparable parts of China, downplays human rights abuses in Xinjiang while questions on Indo-China border problems. This is not a mere byproduct of training data selection but a deliberate function of its design. Liang’s company, deeply embedded in China’s state-directed AI ecosystem, has ensured that DeepSeek is programmed to reinforce official narratives rather than challenge them.

For the CCP, DeepSeek is a powerful new tool in its long-running effort to control public discourse. Traditional censorship methods rely on human moderators and automated keyword filtering, but AI offers a more sophisticated approach. DeepSeek does not simply delete unwanted content; it rewrites history, repackaging state propaganda as AI-generated analysis. The technology can be deployed across state media, social platforms and even education systems, shaping public opinion at scale.

Despite its role as a digital enforcer of state ideology, DeepSeek is no technological backwater. By some metrics, it surpasses Western models in Mandarin-language comprehension, mathematical problem-solving, and industry-specific applications. Chinese companies have been quick to adopt it for enterprise solutions, from customer service bots to automated content moderation. More troublingly, the model is expected to integrate with China’s vast surveillance apparatus, enhancing its ability to monitor and suppress dissent in real time.

Western sanctions have restricted Beijing’s access to advanced semiconductor technology, forcing the country to develop its own AI ecosystem. Liang Wenfeng and his team have managed to train a state-of-the-art LLM despite these constraints, marking a significant step forward for China’s technological ambitions. However, this very success highlights a core contradiction in China’s AI strategy. While DeepSeek is technically advanced, it operates in an environment where knowledge is artificially constrained. AI models thrive on open data and diverse perspectives, yet DeepSeek is deliberately starved of both.

Unlike Huawei’s hardware, which found markets in developing countries, an AI model must be trusted to provide unbiased, accurate information. The model’s limited access to uncensored information presents long-term challenges. DeepSeek will struggle to match the broad capabilities of its Western rivals if it is forced to ignore politically inconvenient facts. True AI innovation requires openness, collaboration, and a willingness to challenge existing ideas - qualities that are incompatible with the CCP’s rigid ideological framework.