China’s Ministry of Industry and Information Technology has issued new network-access rules for smartphones, tablets, wearables and other telecom devices that ship with pre-installed AI agents. The notice was published on November 6 and brings agent-allowd terminals into the country’s existing telecom equipment management framework. It sets expectations for compatibility testing, permission use, communications functions, network and data security, user rights and the way agents execute tasks on a device.
The ministry defines an agent terminal as a telecommunications device with an AI agent pre-installed. The category covers familiar consumer hardware, but the software changes how a device interacts with users and other applications. An agent may interpret an instruction, access information, select a tool and carry out a sequence of actions. That can make a phone more useful, but it also creates risks when the agent requests permissions, handles personal information or performs actions that affect payments, messages and other accounts.
Under the notice, developers are expected to verify that their agents are compatible with specific terminal models and operating systems. The product settings page should disclose a list of devices that have completed the relevant adaptation checks, giving users information about where the agent has been tested. Developers must also submit adaptation information to application distribution platforms before listing an agent. The requirement aims to make compatibility and security status more visible rather than leaving users to discover limitations after installing or activating the software.
Network-access testing will include agent compatibility and fraud risks
The rules also connect agent-allowd devices with network-access testing and anti-fraud assessment. Devices with pre-installed agents must undergo checks that cover agent adaptation and fraud-related risks. That is significant because agents can interact with communication services and may be asked to carry out tasks that resemble legitimate user behaviour. A poorly designed system could expose a user to impersonation, unauthorised actions or manipulation through malicious instructions. The notice places responsibility on developers and device makers to consider these risks before products enter the market.
The regulatory approach is not simply a prohibition on agentic features. The ministry’s accompanying explanation describes the goal as creating a more orderly environment for new terminal products while allowing development to continue. The policy language emphasises that AI should serve users and remain under human control. In practical terms, this points toward a model in which device makers and agent developers must show compatibility and safety, rather than relying only on broad claims that an assistant is intelligent or convenient.
For smartphone manufacturers, the notice may add work to product certification and release schedules. A device that previously needed to show radio and network compliance may now also have to show that its embedded agent behaves appropriately with the operating system and relevant permissions. Developers may need to maintain compatibility lists, document changes and coordinate more closely with app stores. Smaller firms could face a heavier relative burden because they have fewer compliance staff, testing devices and resources for ongoing support.
App permissions and over-the-air updates add compliance obligations
Application developers and platforms are also affected because agents increasingly rely on third-party services to complete tasks. A useful assistant may need to search a marketplace, book transport, send a message or retrieve an invoice. Those interactions raise questions about whether an app permits agent access, what information can be exposed and how a user confirms a consequential action. The notice’s focus on permission management and user rights suggests that agent access cannot be treated as an unrestricted extension of ordinary screen automation.
The most important details for the market will emerge through implementation standards and testing practice. Device makers will need to understand what evidence regulators expect, how agent updates are handled and what happens when a model or operating system changes after certification. Users should watch for clearer compatibility disclosures and permission controls. The notice signals that China intends to regulate AI agents as part of the device and telecom environment in which they operate, not merely as downloadable software.