China's Ministry of Industry and Information Technology, known as MIIT, has published new network-access rules for smartphones, tablets, wearables and other telecom devices that ship with a pre-installed AI agent. The notice, numbered 42 of 2026, was published on 6 November 2026 and brings agent-equipped terminals into the country's existing telecom equipment management framework.

The ministry defines an agent terminal as a telecommunications device with an AI agent pre-installed. Under the notice, those devices must carry an identity code for the agent, in the same framework that governs the licence every phone needs to connect to a network in China. The notice was signed on 24 October and published on 6 November.

What makers must now do

Before a device can receive network-access approval, it must undergo agent compatibility verification under the relevant standards, a fraud-risk security assessment, and advance anti-fraud capability testing. Developers are expected to verify that their agents are compatible with specific terminal models and operating systems, and the product settings page should disclose a list of devices that have completed the relevant adaptation checks, so users can see where the agent has been tested.

The rules also set expectations for how agents use permissions, handle communications functions, protect network and data security, respect user rights, and execute tasks on a device. The aim, as described in the notice, is to make sure agents run stably and that users' rights are protected when software on their phone can act on their behalf.

Why identity codes matter

The identity-code requirement is the most consequential part of the notice. AI agents can send messages, make bookings, move money and contact other services, and a device-level identity makes it possible to trace which agent did what. Analysts writing about the notice have described the codes as passports for bots, a way to give software agents a verifiable identity inside the network rather than letting them act anonymously through a phone.

The move fits a wider pattern in China, where regulators have been tightening control over AI agents as they spread from chatbots into phones and everyday devices. In a separate development reported on 6 November, Chinese authorities moved to restrict state-run enterprises and government agencies from running the agentic platform OpenClaw on office computers, citing security risks from software that holds broad access to private data and can communicate externally.

What this is not

The notice is a departmental measure inside telecommunications network-access management, not a general AI Act. It does not ban AI agents from phones, and it does not set rules for every kind of AI software. Its scope is the device: if a phone, tablet or wearable ships with an agent built in, the maker must prove the agent works safely with that device and can be identified on the network before the device is approved for sale.

Manufacturers selling into China will now need to factor agent testing into their launch plans, including adaptation checks for each model and operating system. How strictly the identity-code and testing requirements are enforced in practice, and whether other countries follow with similar device-level agent rules, will become clear as the first devices go through the new approval process.