A 25-year-old rulebook meets the platform era
On July 3, 2026, the Cyberspace Administration of China (CAC), together with the Ministry of Industry and Information Technology and the Ministry of Public Security, opened a second round of public comment on a revised draft of the Measures for the Administration of Internet Information Services — the administrative regulation that has governed China's internet since State Council Order No. 292 took effect in 2000, back when the country had roughly 20 million internet users. The official notice, also carried by Xinhua, sets comments due by August 2, 2026, via law@cac.gov.cn.
The scale of the rewrite is real: the draft grows from 27 articles to six chapters and 94 articles, folding recommendation algorithms, generative AI, livestreaming and the platform economy — previously scattered across a patchwork of sector-specific rules — into one framework. It is, as the newsletter Geopolitechs put it, China's biggest internet-rulebook rewrite in 25 years.
The steelman: consolidation isn't unreasonable
Regulators have a fair complaint. Since 2000, Beijing has governed AI, algorithms, livestreaming, and platform labor through a scattershot of departmental notices, interim measures and administrative guidance, forcing compliance teams to reverse-engineer a coherent standard from dozens of documents. MLex reports that the new draft explicitly aims to consolidate obligations previously spread across sector rules into a single administrative regulation — a genuine simplification for large platforms trying to build one compliance program instead of ten. The draft's generative-AI chapter also targets problems most democracies are independently legislating against: mandatory labeling of AI-generated content, bans on tampering with those labels, and disclosure of training-data provenance and "basic operating mechanisms" for algorithmic systems — closer in spirit to the EU AI Act's transparency obligations than to blanket censorship. A provision barring platforms from forcing users into AI-driven interactions, and requiring an opt-out from personalized recommendation feeds, is a genuinely pro-consumer instinct, not a uniquely authoritarian one.
Where the draft overreaches
But consolidation is not the whole story. The draft's first chapter states the regulation's guiding principles as upholding Communist Party leadership and "socialist core values" — embedding political loyalty as a legal standard governing every internet information service, not just news publishers. Article-level prohibitions bar algorithms from being used to "disrupt public opinion or evade supervision," a standard vague enough to sweep in ordinary content-ranking decisions a platform makes for engagement, not political reasons. That vagueness is a feature of Chinese platform law, not a bug: it preserves regulators' discretion to act against speech and business practices they dislike after the fact, rather than binding them to a testable rule ex ante.
The enforcement toolkit deepens that discretion rather than narrowing it. The draft creates a "serious breach of trust" blacklist system and empowers regulators to restrict account functions or bar re-registration — administrative sanctions imposed by the same agency that judges the violation, with no indication of independent judicial review built into the draft text. Maximum corporate fines rise from roughly RMB 1 million to RMB 10 million, with individual liability up to RMB 1 million and potential disqualification from executive roles, according to Geopolitechs' reading of the draft. Large platforms — defined by MLex's summary as those with 50 million-plus registered users or 10 million-plus monthly actives, a threshold that captures WeChat, Douyin and Weibo alongside most mid-sized apps — face additional obligations to stand up internal compliance departments and publish social-responsibility disclosures.
Why the distinction matters
The honest comparison isn't the EU's Digital Services Act versus nothing — it's the DSA's appealable, judicially reviewable transparency and risk-assessment regime versus a framework where the same regulator that writes the rule also polices it, blacklists violators, and revises the standard through further "window guidance" outside the text of the Measures itself. A generative-AI labeling requirement is sound policy in Brussels, Washington or Beijing. A labeling requirement married to a discretionary blacklist and a ban on "disrupting public opinion," enforced by the same body that decides what counts as disruption, is a different instrument: one that uses the language of platform accountability to formalize discretionary control over speech and AI products alike.
What to watch
The month-long comment window closing August 2 is unlikely to produce major structural change — China's comment periods on CAC drafts rarely alter core provisions, though they have historically softened specific compliance deadlines. Global platforms and AI developers operating in or selling into China should treat the vague "public opinion" and "supervision" standards as the real compliance risk, not the AI-labeling or algorithm-disclosure provisions, which are broadly consistent with where AI governance is heading everywhere. The test of this rulebook won't be its text but its first enforcement actions after it takes effect.