China's Cyberspace Administration has spent 2026 rewriting the legal foundation for how the internet is regulated in the country. On July 3, the CAC reopened public comment on a revised draft of the Internet Information Service Management Measures — the successor to a regulation that has governed Chinese internet platforms since 2000 — and the comment window closed August 2. The new draft runs to six chapters and 94 articles, nearly double the 54-article version that circulated for comment in 2021 and more than triple the original 27-article 2000 rules, according to an analysis of the filing by the China-policy newsletter Geopolitechs.
The headline addition is an entirely new chapter on "Intelligent Information Services" — Articles 55 through 64 — which for the first time folds algorithm recommendation, generative AI, and AI agents into the core internet-services law rather than leaving them in standalone departmental rules. Article 61 is the operative provision for recommender systems. It requires platforms to build "traffic rules and algorithmic standards with positive value-guiding functions" (具有积极价值引导功能的流量规则和算法标准) into their ranking logic; it bans using algorithms to block information, over-recommend content, manipulate rankings or search results, or control trending and featured lists; it prohibits forcing users to use personalized recommendation at all; and it requires platforms to offer either a non-personalized option or "a convenient means to refuse" algorithmic recommendation.
This Is Not Entirely New
Much of Article 61 is a restatement, not an invention. China's dedicated Algorithm Recommendation Provisions have been in force since March 1, 2022, and already required platforms to "promote positive energy" (Article 6) and let users switch off personalized recommendation entirely, with immediate effect once they do (Article 17). What Article 61 does is elevate those obligations out of a joint CAC-MIIT-Ministry of Public Security departmental rule and write them into the successor to the foundational internet-services measure — the law that, in its original 2000 form, has anchored China's entire internet-licensing regime. That is a change in legal weight, not just wording.
The Case for the Rule
The strongest argument for Article 61 isn't censorship — it's consumer protection, and it deserves to be taken on its own terms. Recommendation algorithms genuinely can be engineered to manufacture artificial urgency: manipulated trending lists, astroturfed search results, and "forced push" personalization that users can't turn off are real dark patterns, and they exist on Chinese platforms for the same commercial reasons they exist everywhere — engagement optimization pays, transparency doesn't. A hard-coded opt-out with an "immediate effect" requirement, as the 2022 rules already establish and Article 61 reaffirms, is a genuinely pro-consumer standard; U.S. and EU platforms offer far weaker versions of the same idea, usually behind several settings menus rather than a single switch. Banning trending-list manipulation specifically addresses a documented problem: China's own state media have repeatedly reported on paid "hot search" placement schemes on Weibo and Douyin as a real commercial-manipulation issue, not a manufactured one.
Where the Rule Overreaches
The problem is that Article 61 bundles that legitimate consumer-protection logic with an ideological content-steering mandate, and gives regulators no way to tell platforms apart on the two grounds. "Positive value guidance" is not a technical standard — it has no measurable definition, no safe harbor, and no independent adjudication. A platform cannot audit its ranking algorithm against a requirement to be "positive"; it can only guess what regulators will accept after the fact, which is precisely the compliance environment that produces defensive over-moderation. That dynamic already shaped the 2022 rules' rollout: platforms facing an undefined "positive energy" standard tend to suppress broad categories of legitimate commentary, satire, and minority-opinion content rather than risk enforcement, because under-blocking carries regulatory risk while over-blocking carries none. Folding this into a higher-status, harder-to-litigate law only raises the stakes of that same asymmetry.
There's also a real compliance-cost question for an industry the government says it wants to keep globally competitive. Chinese platform companies — Douyin's parent ByteDance, Tencent, Alibaba, Kuaishou — already operate under an algorithm-filing regime, a separate generative-AI labeling rule, and now a third overlapping standard for "value-guiding" ranking logic embedded in the base internet law. Each new layer adds legal-review cost without a correspondingly narrow definition of what it prohibits, which disproportionately burdens smaller platforms that lack in-house teams to interpret vague standards the way the largest incumbents can.
What to Watch
The published draft leaves its own effective date blank, and CAC has not indicated a target for finalization. The provisions worth tracking separately are the ones with actual technical content: the ban on trending-list manipulation and the opt-out mandate are enforceable, auditable, and defensible as ordinary platform-integrity rules that any regulator — Chinese, European, or American — could plausibly adopt. The "positive value guidance" language is the one that deserves scrutiny, because it is the one that converts a content-neutral technical standard into a discretionary speech lever, dressed in the same paragraph as rules that have nothing to do with speech at all.