Overview
The Ministry of Industry and Information Technology recently issued a notice on strengthening network-access management for agent-equipped terminal devices, establishing an end-to-end framework—deployment adaptation verification, network-access testing, over-the-air (OTA) update review by experts, and app-distribution disclosure—for phones, tablets and wearables with pre-installed agents. It explicitly curbs forced agent invocation and incompatible installations to protect users’ right to know and choose. This is China’s first systematic network-access framework for the “agent terminal” as a new category.
Around the same time, the Cyberspace Administration launched a “Clear” campaign against AI-generated health misinformation, requiring platforms to improve detection and handling of synthetic health rumors. Together the moves signal a shift in AI governance from “model content” to “terminals, scenarios and behavior.”
Background & Interpretation
1. Technical Background and Evolution
Regulation previously targeted training data, model filing and content labeling. As agents move from cloud to phones, vehicles and appliances, AI gains the ability to “autonomously call tools and act on the user’s behalf,” raising risks from “saying wrong things” to “doing wrong things.” The terminal is the first interface between AI, the physical world and personal data, demanding front-end rules.
2. Drivers and Underlying Mechanism
The notice’s logic is “liability shifting forward”: embedding safety and compliance into network-access and OTA stages makes device makers verifiably responsible for pre-installed agent behavior, while banning forced invocation and ensuring compatibility protects user choice and prevents de-facto monopolies via system privileges. The CAC campaign blocks the social harm of AI health rumors from the content side.
3. Market and Industry Implications
The rules raise compliance costs for agent-device makers but build trust barriers for early compliers. Phone, wearable and appliance makers must rebuild adaptation, testing and upgrade flows for pre-installed agents; security testing, compliance consulting and on-device AI governance toolchains see new demand. Globally, Meta’s pause of contested scraping terms and EU AI Act enforcement contrast “democratic front-loading” with “access gatekeeping.”
Implications & Outlook
Near term, device makers will rush to complete network-access testing and OTA review processes. Medium term, “agent-device security certification” may become a new 准入 label alongside 3C, spawning an on-device AI governance niche. Key variables: rollout of implementing rules and testing standards, makers’ compliance progress, and consumer acceptance of agent devices.
Takeaways for Industry Participants
Device makers should front-load compliance into product definition, building agent-behavior auditing and revocation mechanisms. App developers must respect authorization boundaries, avoiding forced invocation and data overreach. Enterprises procuring on-device AI should treat network-access compliance and security certification as hard selection criteria.
This column compiles industry information and shares technical perspectives; it does not constitute investment advice.