A Chinese actress named Wen Zhengrong logged on one day in late 2025 and found herself selling three different products in three different livestreams at the same time, wearing three different outfits. However, she had filmed none of it. Every version was an AI deepfake of her face and voice. When she confronted one of the accounts and asked, “if you are Wen Zhengrong, then who am I?”, she was blocked. Days later, the Cyberspace Administration of China said it had dealt with a batch of accounts that used AI to mimic celebrities in product livestreams.
That episode tells you most of what you need to know about how China governs artificial intelligence. There’s no grand “AI Act.” There’s a watchdog, a stack of narrow rules, and a willingness to act fast when something goes wrong.
Does China Have an AI Law? No, and That’s the Strategy
Ask people how the European Union handles AI, and they’ll point to one document: the EU AI Act. Ask the same about China, and there’s no single answer, because no single law exists.
China does not yet have a standalone, comprehensive AI statute. Instead, Beijing controls the technology through a series of smaller, targeted rules, each aimed at a specific problem. Recommendation algorithms got their own rulebook. Deepfakes got another. Chatbots got a third. Labeling came later. Scholars proposed a draft national AI Law in 2024, and lawmakers floated the idea again in mid-2025, but as of June 2026, it remains a proposal rather than a law.
The piecemeal approach lets regulators write a rule, watch how the market reacts, and write the next one based on what they learned. The downside is obvious for any company trying to comply: you’re tracking a moving target made of many parts, not reading one master text.
The CAC Is the Referee, and It Almost Never Works Alone
If China’s AI rules have a center of gravity, it’s the Cyberspace Administration of China, or CAC. Several bodies share oversight, but the CAC is the lead authority. It started in 2014 as an internet content regulator and has since expanded into cybersecurity, data policy, and now AI.
The CAC rarely signs a rule by itself. The big AI regulations carry several ministry names at once, like industry and information technology, public security, science and technology, and others. That joint sign-off matters. A rule arrives with buy-in from the agencies that will enforce it.
The Deepfake Rule Came First, Back in January 2023
China was early on deepfakes. The Administrative Provisions on Deep Synthesis in Internet-based Information Services took effect on January 10, 2023. “Deep synthesis” is the official term for what most people call deepfakes, like AI-made or AI-altered video, audio, images, and text.
The rule does a few plain things. Anyone making realistic fakes has to label them. They need consent before recreating a real person’s face or voice, and providers whose tools can sway public opinion must register with authorities, a system that requires real paperwork. The CAC has logged over 2,800 deep synthesis algorithms across nine filing batches.
Beijing frames deepfakes as a security matter, not just a consumer nuisance. In late 2025, the Ministry of State Security warned that the technology could endanger individual rights, social stability, and national security, claiming some hostile foreign forces had used fake videos to mislead opinion inside China.
Chatbots Live Under the 2023 Generative AI Rules
When tools like ChatGPT set off a global scramble, China answered within months. The Interim Measures for the Management of Generative AI Services took effect on August 15, 2023, as the first regulation specifically designed for generative AI. If you offer a chatbot or image generator to the public in China, these are the rules you follow.
Providers must use lawfully sourced training data, respect intellectual property and personal information, and improve the accuracy of what their models produce. Anyone with public-opinion influence must pass a security assessment and file with regulators before launch.
The measures require providers to uphold “socialist core values” and bar content that endangers national security or incites the overthrow of the system. Penalties are real, with fines up to 50 million RMB or 5% of annual revenue under the Personal Information Protection Law. These rules reach beyond China’s borders. They apply to anyone serving the public in China, regardless of where the company is based.
Since September 2025, AI Content Must Wear a Label
The newest big rule answers a simple question: how do you know what’s real? On March 14, 2025, the CAC released the final Measures for Labeling AI-Generated Content, which took effect on September 1, 2025.
The rule sets up two kinds of labels. Explicit ones are what you can see or hear, like the small tags on text, audio, images, and video that tell you a machine made this. Implicit ones sit inside the file’s metadata, which are invisible to the eye but readable by machines. Tencent, Douyin, RedNote, Weibo, and DeepSeek all issued public compliance statements.
China’s Humanoid Robot Digital ID and AI Companion Rules
What separates a 2026 view from a 2025 one is where regulators now point: machines that act human.
In late May 2026, China rolled out a national “digital ID” system for humanoid robots, giving each machine a 29-digit code through an MIIT-backed platform. The code breaks into four parts: a country code, an enterprise code, a product model code, and a serial number. By launch, more than 28,000 robots across 200 models were already registered, and the rule is blunt: no code, no domestic-market access.
The system made the rounds on X, where former UK politician Andrew Bridgen framed it as Beijing making sure robots “won’t go rogue.” The numbers are accurate, but the framing oversells it. The official purpose is traceability and liability, a clear paper trail connecting a machine to its maker if a robot injures someone, modeled on the cradle-to-grave tracking used for aircraft, not a behavioral leash.
The same instinct shows up in software. The CAC has circulated draft rules for human-like interactive AI services, covering AI companions and chatbots, with a comment window that closed in January 2026. A second draft targets “digital humans”, AI built to mimic emotional intimacy, with children flagged as a special concern. The pattern holds: a new use of AI appears, the CAC writes a rule for that specific use, and the rulebook grows by one more entry.
What China’s AI Regulation Means for Companies in 2026
China governs AI the way you’d assemble a toolkit rather than write a constitution. The CAC and its partner ministries write a rule for each problem as it surfaces: deepfakes and chatbots in 2023, labeling in 2025, robots and AI companions now in 2026. There’s no single law to read, and that’s the point. The system trades the clarity of one master text for the speed of many small ones.
For companies, the takeaway is that you don’t comply with “China’s AI law,” because there isn’t one. You comply with whichever rules your product touches, and you watch the CAC’s drafts closely, because the next rule is usually already in consultation.
Europe is betting on the opposite model, one comprehensive framework meant to last. The critical unknowns are whether this approach will prove effective and whether the EU can act with sufficient speed to maintain its competitive edge in AI.
Author: Ayanfe Fakunle
See Also:
Will the EU’s AI Act Cripple Europe’s Innovation Edge?
EU AI Act August 2026 Deadline: What Startups Need to Know
What is the European Innovation Act? How the EU is Bridging Europe’s Research-to-Market Gap
