China issues first network-access rules for AI-agent devices

MIIT's new rules are the first network-access regime written specifically for consumer devices with built-in AI agents. They cover phones, tablets and wearables. Before a device can be connected to Chinese networks, it must pass adaptation testing and anti-fraud testing. Each device must also carry a unique identity code, which can be used to trace actions taken by its agent.

The rules follow how agents actually behave on phones. An agent can tap through apps, fill in forms and complete payments on the user's behalf. That creates fraud and accountability risks that older device certification never considered. Tying agent actions to a device identity gives regulators and platforms a way to attribute those actions.

The rules land as Apple begins its China trial of Apple Intelligence running on Alibaba's Qwen, with public access planned for December. Domestic handset makers are also racing to ship agentic assistants. Any vendor selling an agent phone in China now has a compliance checklist to meet.

What to watch: implementation timelines, and whether third-party agent apps fall under the rules alongside pre-installed agents. Also watch how strictly the anti-fraud testing is enforced against foreign vendors.