Tang Jie, co-founder and chief scientist of China’s leading artificial intelligence laboratory Z.ai, embodies the country’s distinct approach to AI development and safety. His leadership reflects a vision of AI safety that prioritizes strict governance aligned with the Chinese Communist Party’s political security, contrasting with Western discussions centered on societal risks.

Z.ai, headquartered in Beijing and affiliated with Tsinghua University’s Knowledge Engineering Group, has established itself as a top-tier AI lab in China. Mr. Tang has asserted that the immense power of AI demands "extreme safety governance," framing safety not as an afterthought but as a prerequisite for AI’s survival and use. His position echoes concerns from American AI experts but diverges sharply in focus, emphasizing stability and control in service of China’s political system.

This difference in perspective is rooted in the Chinese government’s interpretation of AI safety. At the World Artificial Intelligence Conference in Shanghai in mid-July, President Xi Jinping underscored China’s commitment to ensuring AI is “safe, secure and controllable.” However, official translations of Xi’s remarks differ, with the original language highlighting “development and security” and emphasizing control tied explicitly to the Communist Party’s political endurance.

Chinese AI developers, including Mr. Tang and his colleagues, incorporate Communist Party values deeply into AI models. These systems are designed to filter politically sensitive content and align with party orthodoxy, reflecting state censorship and ideological priorities. This approach contrasts with Western AI principles, which tend to emphasize universal human rights and individual liberties.

For example, companies like Anthropic and DeepMind, based in the United States and the United Kingdom respectively, embed safety concepts inspired by liberal democratic values such as freedom, equality, and respect for human rights into their AI frameworks. Anthony Amodei, Anthropic’s CEO, likens AI safety to teaching a child ethical principles, referencing the Universal Declaration of Human Rights as a foundation. Google DeepMind similarly advocates for democracies to lead AI development guided by these core values.

In contrast, China’s AI safety framework reflects concerns about external threats to state power and socialist values. Chen Yixin, Minister for State Security, recently emphasized the need to defend against hostile forces that might exploit AI for “direct assault” on China’s political and ideological security. His focus is less on AI turning against humanity and more on AI’s role in protecting the party’s governance and national sovereignty. China’s evolving regulations explicitly aim to safeguard state power and political stability as part of AI safety.

The divergence in understanding AI risks has complicated efforts to establish cooperative international frameworks. Ahead of a scheduled meeting between President Trump and President Xi at the White House, observers noted that the two nations’ contrasting definitions of AI safety would make finding common ground challenging.

Mr. Tang’s letter to Z.ai staff in July encapsulated the Chinese narrative, rejecting the idea of applying “bolt-on” safety measures and instead advocating for embedding AI behavioral principles from the outset. As Chinese AI systems are adopted worldwide, there is growing attention on how their underlying values, shaped by the Communist Party’s worldview, might influence global AI norms and governance.