AI Governance in the US vs. China
Why China is winning.
Introduction
The battle for global leadership in AI is often framed as a competition about the raw capabilities of models. The US is a clear winner, because no other country has the ability or willingness to throw as much venture capital and public funding behind datacenters and expensive training runs.
I will argue that governance is even more important to ‘win’ in the AI economy. ‘AI governance’ means establishing the systems, laws, policies, human resources, organizational and institutional capacities to prevent and absorb the negative impacts of AI, while harnessing its potential. Without the proper governance frameworks and oversight mechanisms there is no way to safely deploy the technology at scale.
AI is categorically different from past tech revolutions whether cars, aviation, pharmaceuticals, renewable energy, personal computers, or the internet itself. As Henry Farrell et al. puts it “Large AI models are cultural and social technologies” which means that their impacts are not limited to economics or people’s lifestyle choices. Large AI models will also affect how we relate to society and each other on a fundamental level. Another example of a “’cultural and social technology’ is social media, which the US, EU member states and most other countries have utterly failed to govern, while China was more prescient about its dangers.
China is currently positioned to become a world-leader in AI governance, while the US is positioned to lose big time by repeating the same mistakes it committed with social media. Instead of governing the technology for the public good, private companies are essentially allowed to do what they want, which means that a few companies grow big and powerful, while the population, institutions, and other businesses are left poorer in both spirit and pockets. Europe remains mostly a testing ground for American technology without much self-determination or autonomy, until it develops a separate identity apart from the US and a digital infrastructure of its own – despite its regulation.
AI Governance in China vs. the US
Interim Measures for the Administration of Generative Artificial Intelligence Services entered into force in July, 2023. The law strictly regulates how GenAI services can operate in China. For example, GenAI services are required to adhere to the core socialist values, take effective measures to prevent discrimination, respect copyright laws, business ethics, and the legitimate rights and interests of others. GenAI services with “public opinion attributes or social mobilization capabilities” are required to register different types of information about the system in a public database (same procedure as with recommendation algorithms) and undergo a government security assessment, before they can be offered to the public. China has also adopted comprehensive laws on labelling of AI-generated content and deepfakes.
In western countries, AI girlfriends have become notably popular among young men. In China, it’s primarily the other way around as adult women seek to meet their romantic needs through AI boyfriends. The cultural perspective is interesting here, but that is not why I bring it up. The reason is that China is the first country to my knowledge which has presented a comprehensive and targeted regulation of the emerging and strange market for AI companions.
On December 27, 2025, the Cyberspace Administration of China Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interactive Services for public solicitation. The law is set to enter into force already in July, 2026, and applies to AI technology which “simulate human personality traits, thinking patterns and communication styles and interact emotionally with humans through text, pictures, audio, video, etc.” Among other requirements to these AI systems with anthropomorphic qualities, they are not allowed to:
Generate or disseminate content that promotes obscenity, gambling, violence, or incitement to crime.
Provide false promises that seriously affect user behavior and services that damage social and interpersonal relationships.
Harm users’ physical health by encouraging, glorifying, or suggesting suicide or self-harm, or harming users’ personal dignity and mental health through verbal violence or emotional manipulation.
Induce users to make unreasonable decisions through algorithmic manipulation, misleading information, and setting emotional traps.
Induce or extract confidential and sensitive information.
Article 11 of the law states:
“Providers should have the ability to identify user status. On the premise of protecting users’ personal privacy, they should assess users’ emotions and their dependence on products and services. If they find that users have extreme emotions and addiction, they should take necessary measures to intervene.”
Using such vague language to describe people’s inner states is not usually a feature of laws in Western democracies which evolve around observable behaviors and actions. The problem is that the harms of AI companions are rarely observable until it’s too late and the user has developed an “AI psychosis”.
China’s laws set out strict and numerous requirements and obligations to AI companies. This approach makes complete sense, since only the companies that develop and provide AI systems to the public are in a strong position to mitigate the harms and risks before they are realized. Let’s contrast these governance measures with AI laws in the US.
In the US, the regulation of AI is currently fragmented among individual states. There is no federal framework to address the risks and impacts of AI and the government is actively opposing “onerous” state laws. However, the White House recently published a series of legislative recommendations to the US Congress in a document titled A National Policy Framework for Artificial Intelligence. It proposes that Congress must establish a Federal AI Policy Framework that preempts state laws and shall include certain measures to protect children from harm online, local communities from increased energy costs due to datacenters, creators, publishers, and innovators from copyright infringements, while also defending “free speech” and “removing barriers to innovation”. What is missing from the proposed framework is any obligations or requirements directed at AI companies.
Big Tech influences policy-making in the US to such an extent that one starts to wonder who really is in control. The companies assert influence through a record-breaking $109 million in lobbying, Super PACs, lobby-drafted legislation, charm offensives with lawmakers, and warnings about its powerful technology. Above all, Big Tech and AI have become such a foundational pillar of the US economy, that democratic governing powers have little choice, but to obey the industry’s demands.
That is why AI governance is failing in the US and succeeding in China. Downplaying societal harms and risks lead to more profits and economic growth in the short term, but the lack of central oversight and governance undermines trust in institutions and lead to complex and unfavorable outcomes in the longer term.
