While the West debates system cards and model safety, Beijing is creating the TCP/IP of autonomous AI — and may determine who talks to whom in the next era of computing.
There is a curious asymmetry in how the world is handling AI regulation. In the West, we spend our energy debating the safety of language models — training transparency, system cards, existential risks. China, meanwhile, is solving a different and possibly more relevant problem: how autonomous AI agents will know who they are, find each other, and communicate. In May 2026, Beijing's SAC/TC 28 technical committee published GB/Z 185, a seven-part national standard covering identity codes, identity management, description, discovery, interaction, and tool invocation for AI agents. According to an analysis by researcher Ken Huang, this is a technical standard for agent interconnection. In other words: China is not regulating the brain; it is standardizing the nervous system.
This is a philosophical divergence with geopolitical consequences. The West treats AI as a content problem — what the model says, what it can generate, what risks it poses. China treats AI as an infrastructure problem — how these systems connect, what protocols they use, who controls the exchange points. The first approach produces ethics committees and policy documents. The second produces technical standards that, once adopted at scale, become nearly impossible to undo.
The obvious analogy is with the internet of the 1980s and 1990s. TCP/IP did not win because it was technically superior to alternatives like OSI. It won because it was adopted early, openly, and broadly, and because it created network effects that rendered any competing protocol irrelevant. GB/Z 185 is a bet on the same kind of path dependency. If Chinese AI agents learn to identify, discover, and interact using a national standard, that standard becomes the lingua franca of autonomous AI in one of the world's largest digital markets. And standards, once established, tend to spill across borders.
Some argue that the West is already working on agent communication protocols — Anthropic's MCP, Google's A2A, and initiatives from various startups. This is true. But there is a difference between a corporate protocol and a national standard. The former is a product strategy; the latter is a state strategy. China has a consistent track record of using standardization as a tool for projecting power: it did so with 5G, mobile payments, and QR codes. In each case, the domestic standard became a platform for global influence. With AI agents, the bet is even more ambitious, because whoever defines how agents communicate ultimately dictates which AI ecosystem operates on a planetary scale.
The West's blind spot is treating this as a minor technical issue. It is not. The identity and interconnection layer for agents is where the real power of autonomous AI will consolidate. A language model is an engine; an agent protocol is the road. You can have the best engine in the world, but if the roads are all built to someone else's standard, you drive on their terms. And China understood this before most people even realized the road was being built.
The West still has time to articulate a response — but not through yet another system card. The response must be an open, multilateral, and technically robust standard for agent identity and interconnection, before GB/Z 185 becomes the default protocol in half the world. The question is not whether AI will be regulated. It is at which layer regulation takes place — and who controls it.
Published by Beijing's SAC/TC 28 technical committee, GB/Z 185 is a seven-part national standard that covers identity codes, identity management, description, discovery, interaction, and tool invocation for AI agents. It serves as a technical standard for agent interconnection.
The West treats AI primarily as a content problem, focusing on model safety, training transparency, and system cards. China treats AI as an infrastructure problem, focusing on standardizing how autonomous AI agents connect, identify each other, and communicate.
Standardizing protocols creates path dependency and network effects, similar to how TCP/IP became the default internet protocol. Whoever controls the communication and identity layer for agents dictates which AI ecosystem operates globally, making infrastructure standards more strategically powerful than regulating individual models.