--- title: "China's foundational internet law gets an AI clause — but it's still just a draft" date: 2026-10-31 category: Policy & Governance site: NeuroAI canonical: https://neuroai.site/a/na-policy-internet-info-smart-service-draft language: en --- # China's foundational internet law gets an AI clause — but it's still just a draft > China's Internet Information Service Measures revision draft adds a 'smart information service' chapter on AI labeling and algorithms — still a draft, not in force. The regulation that has governed China's internet since the mid-1990s is being rewritten, and artificial intelligence is the reason. For the first time, a foundational administrative rule would define a "smart information service" (智能信息服务) and set obligations for anyone using AI to publish information online. But readers should not mistake momentum for law. The text is a revision draft open for public comment (修订草案征求意见稿) — not yet adopted, and not in force. ## Key takeaways - The Cyberspace Administration of China (CAC, 国家互联网信息办公室) republished a revision draft of the *Internet Information Service Management Measures* for comment on 3 July 2026. - The draft's headline addition is a new "smart information service" (智能信息服务) chapter, the first time this concept appears at the administrative-regulation level. - It would require AI synthetic-content labeling (生成合成内容标识), algorithm transparency, training-data compliance and algorithm-filing for opinion-shaped services. - The public-comment window closed on 2 August 2026; the text remains a draft and has not been adopted or taken effect. - It would lift existing rules — the algorithm-recommendation rule (算法推荐管理规定), deep-synthesis rule and generative-AI interim measures — from departmental rules to a higher legal tier. - The draft also adds chapters on platform responsibility, minor protection and labor rights for platform workers. ## The law and the new clause The *Internet Information Service Management Measures* are a foundational administrative regulation that has been in force for roughly 26 years. The revision, running six chapters and 94 articles, keeps the old scaffolding but inserts dedicated chapters for modern problems. The most watched is the "smart information service" (智能信息服务) chapter, which defines a "smart information service provider" (智能信息服务提供者) as any entity using AI to provide internet information services. ## What the chapter requires The chapter translates existing practice into firmer language. Providers must publicly disclose, as appropriate, the basic principles, purpose, main operating mechanism and training-data sources (训练数据来源) of their technology — a direct answer to the "algorithm black box" complaint. Services with opinion-shaping or social-mobilization attributes must run security assessments and complete algorithm filing (算法备案). On content, providers offering generation must label AI synthetic content (生成合成内容标识) under the mandatory national standard, and no organization or individual may maliciously delete, alter, forge or hide such labels. On algorithms, recommendation services may not force users to use smart information services, must offer non-personalized options, and may not use algorithms to manipulate rankings, hot-search lists or search results to disturb public opinion or evade supervision. A separate article extends labor protections to platform workers whose scheduling, pay and hours are set by algorithms. ## Why it matters: from rules to law Until now, China's AI governance has rested on a stack of departmental rules issued piece by piece — the algorithm-recommendation rule (算法推荐管理规定) in 2022, the deep-synthesis rule, and the generative-AI interim measures in 2023. Each was useful but narrow, and none carried the weight of an administrative regulation. Pulling them into a single "smart information service" (智能信息服务) chapter at the regulation tier is meant to create one coherent, higher-status framework — what legal scholars describe as "adaptive governance" that keeps flexible escape hatches for fast-moving technology. The AI chapter is not the only change. The same draft tightens platform accountability through a new "platform information service" chapter covering public accounts, livestreaming and multi-channel distribution agencies, and it strengthens minor-protection and anti-cyberbullying duties. Read together, the revision signals a broader tightening of internet governance, with the smart-information-service rules as the most novel piece rather than the whole story. For foreign firms, the draft's reach also extends to overseas organizations and individuals providing such services — a point worth watching as the text moves toward adoption. ## Still a draft — read the stage carefully This is the part most often misreported. The CAC notice of 3 July 2026 was a *second* round of public comment ("再次公开征求意见"), and the window closed on 2 August 2026. As of late October 2026 the text is a revision draft (修订草案征求意见稿) only. It has not been passed by the State Council, has no effective date, and its articles can still change. Anyone treating it as binding — quoting fines, deadlines or obligations as if live — is wrong. The correct status is: proposed, commented on, not yet law. ## Honest limitations This article draws on the CAC's official comment notice, the accompanying explanatory note, and legal-affairs and state media analysis; I have not compared every article against the 1994 original line by line. The "first time at regulation tier" claim refers to the specific concept of a "smart information service" chapter, not to all AI obligations, some of which already exist in lower-tier rules. I have not verified whether a follow-up adoption timeline has been announced. The scholarly "adaptive governance" framing is analysts' interpretation, not the law's text. ## What readers can do now - If you operate an AI information service in China, comply now with the *existing* rules the draft would consolidate — labeling, algorithm filing and training-data obligations are already live. - If you are a platform, prepare for the draft's labor and transparency articles by auditing how algorithms set worker schedules and how you disclose model mechanisms. - If you follow policy, track the State Council's legislative calendar for the adoption step, and do not treat the draft's penalties or dates as enforceable until then. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-policy-internet-info-smart-service-draft Free to quote with attribution and a link to the original.