On 3 July 2026 the Cyberspace Administration of China (CAC) re-opened public consultation on the revised draft of the Measures for the Administration of Internet Information Services, with comments due by 2 August 2026. The Measures were last revised in 2000 — before mobile internet, before algorithmic recommendation, and certainly before large models. The structural arrangement of this draft is worth noting: it adds two dedicated sections, “platform information services” and “intelligent information services”. For companies providing AI information services, this means artificial intelligence is no longer merely swept up by general internet information service rules; it has a rulebook of its own.
Why a 26-year-old regulation is being revised
The current Measures took effect in 2000 and established the basic framework of operating licences, filing and content management, which has long been applied unchanged. In those 26 years the shape of the internet changed fundamentally: information services moved from portals to mobile apps, from human editing to algorithmic distribution, and from static content to generative content.
The explanatory note to the revision is direct — the draft was amended and improved in light of the amended Cybersecurity Law and other laws, and of public feedback. The overall approach is “refined governance”: taking institutions already established in higher-level law and refining them downward, and elevating practices that have proven effective into regulation. Expert commentary further notes that the revision focuses on the primary responsibility of network platforms and the new challenges that AI technology brings to information-content governance.
In other words, the revision answers two questions: what responsibility platforms should bear, and what new variables AI introduces into content governance.
Two new sections: platform information services and intelligent information services
The draft creates dedicated sections for “platform information services” and “intelligent information services” within its operational chapter. The former targets operators providing platform services; the latter targets “the provision of internet information services using artificial intelligence technology”.
Singling out AI information services has a point: regulatory precision. The duties of general internet information services are designed around the chain of “publish — disseminate — handle”. The core of AI information services, by contrast, is model generation and algorithmic distribution, where the sources of risk and the available control points are different. A dedicated section is an acknowledgement that these services need their own compliance actions rather than a simple extension of content moderation.
Four institutional points in the intelligent information services section
According to the explanatory note, the intelligent information services section mainly clarifies four areas.
First, generative and synthesised content labelling. It provides for the labelling of AI-generated and synthesised content in accordance with law. This dovetails with the already-effective Measures for the Labelling of AI-Generated and Synthesised Content — labelling is lifted from the level of a departmental rule to a duty at the level of an administrative regulation.
Second, constraints on the use of algorithmic recommendation technology. The section sets two prohibitions: users must not be forced to use intelligent information services; and algorithms must not be used to disrupt public opinion or evade supervision. The former targets “on by default” or “cannot be turned off” product design; the latter targets algorithms used to interfere with the order in which information is presented.
Third, protection of workers’ rights, directly linked to platform information services. This is a provision that is easily overlooked but has direct impact on platform companies. Algorithmic dispatch, AI scheduling and automated assessment act directly on workers’ work arrangements and pay, and such relationships require corresponding rights-protection arrangements.
Fourth, strengthening responses to significant AI security risks. For AI information services that may produce significant security risks, corresponding response mechanisms are required. This is consistent with the approach to risk grading and technical response in China’s AI Safety Governance Framework 3.0, which we discuss in our EU AI Act and cross-border compliance practice.
Platform-side baseline duties are tightening in parallel
Beyond the dedicated sections, the general provisions of the draft thicken existing duties, several of which bear directly on AI information service companies.
On account management, Article 19 requires real-identity verification of users; Article 21 establishes a dynamic account-verification system under which accounts inactive for more than six months may be prompted, have functions limited, or be frozen, cancelled or barred from re-registration in accordance with law and agreement. Article 20 specifically prohibits evading verification through false identity, impersonation or illegally obtaining another’s account.
On content governance, Article 24 lists prohibited categories of information; Article 25 requires the prevention and resistance of harmful information such as hype around emergencies and content that may affect the physical and mental health of minors; Article 31 requires a rumour-refutation mechanism. Article 26 brings “operational” acts — publishing, deleting, blocking, disconnecting links, replacing, down-ranking and algorithmic recommendation — into scope, and prohibits traffic fraud, manufacturing fake opinion hotspots, and manipulating rankings and trending lists.
On service safeguards, Article 34 requires a complaints, reports and appeals system and human customer service matched to business scale; Articles 35 and 36 set requirements for minors mode and for age-friendly and accessible services. Taken together, these provisions form a more detailed operating compliance baseline than the current Measures.
What it means for AI companies in practice
For companies providing AI information services, the draft has three layers of meaning.
First, labelling is a hard constraint that must be delivered at product level. The labelling of generated and synthesised content is not a remedial step after the fact; it must be built into generation, editing and dissemination. For existing discussion of labelling and content governance, see our Algorithm Filing service page.
Second, the compliance focus of algorithmic recommendation expands from “whether it is filed” to “how it is used”. The prohibitions on forced use and on disrupting public opinion point to default settings and the design of the algorithm’s objective function, and must be assessed at the feature-design stage. See our Algorithm Filing service page for filing and assessment requirements.
Third, the workers’ rights clause signals that platform-type AI services need to revisit their employment arrangements. Where algorithms directly determine dispatch, hours and assessment, corresponding rights protection and compliance arrangements must be aligned with labour-law requirements.
The draft is still at the consultation stage, and the final text may differ. But the direction is clear: compliance for AI information services is expanding from a single point — content moderation — into a composite set of requirements covering labelling, algorithms, employment and security risk. Companies can use this to review product features and internal systems in advance and make targeted adjustments once the final text is issued.
This article is based on the revised draft of the Measures for the Administration of Internet Information Services published for consultation by the Cyberspace Administration of China on 3 July 2026, its explanatory note, and public expert commentary. It is for reference only and does not constitute legal advice. Specific compliance judgements should be based on the officially effective text and applicable law.