China's interim measures made AI providers police overuse

The Interim Measures for generative AI services set content, filing and minor over-reliance duties, effective 15 August 2023.

Historical event

August 15, 2023

First source published

July 10, 2023

Site publication

September 18, 2026

What happened

China's 生成式人工智能服务管理暂行办法 ('Interim Measures for the Management of Generative Artificial Intelligence Services') were approved at the Cyberspace Administration of China's twelfth 2023 office session on 23 May 2023, jointly issued with six other ministries on 10 July 2023, and took effect on 15 August 2023 — adoption and application are two separate dates in the text itself. The original text, published on the Cyberspace Administration's own site, and China Law Translate's English rendering agree on twenty-four articles covering any service that uses generative AI to produce text, image, audio or video content for the public inside mainland China.

What the documents show

Article 4 requires providers to prevent generated content that incites subversion, undermines national unity, promotes terrorism or ethnic hatred, or is violent or obscene, and to take effective measures against algorithmic discrimination. Article 9 makes a provider responsible as the producer of the content it generates and as the handler of any personal information involved, and requires a signed service agreement with registered users. Article 10, the provision most directly relevant to companion-style products, requires a provider to disclose its intended user group, occasions and uses, and to employ effective measures to prevent minor users from over-reliance or addiction to generative AI services (防范未成年人用户过度依赖或者沉迷生成式人工智能服务). Article 11 bars a provider from illegally retaining or disclosing identifiable user input and requires it to honour requests to access, correct or delete personal information.

The mechanism

The Measures work through a producer-liability model rather than a pre-market licensing gate for most services: a provider bears the same legal responsibility for AI-generated content that it would for human-authored content it published, plus AI-specific duties layered on top, including content labelling under separate deep-synthesis rules, security assessment and algorithm filing for services with public-opinion or social-mobilisation capacity, and the minor-overuse duty in Article 10. That duty is a general obligation to take effective measures; the text does not specify what those measures must be, leaving the standard to sector guidance and enforcement.

What it leaves open

Article 10's addiction-prevention language does not define a compliance benchmark, no mandated time limits, no specified minor mode, so what counts as an effective measure against over-reliance is left to regulators and, in practice, to platform-specific implementation. The Measures also apply only to services offered to the mainland Chinese public, so they say nothing about how a China-based provider's export products are governed. This is an editorial point: the same Article 10 duty referenced for general chatbots would, on its plain text, apply equally to a companion-style app operating in China, without a companion-specific carve-out or addition.

• What specific design change did a provider make to satisfy Article 10's duty against minor over-reliance?
• Is the service registered and filed as required for algorithms with public-opinion or social-mobilisation capacity?
• Does the provider's own stated user group and intended use match how the product is actually marketed?

The Measures pair a broad content-responsibility rule with one explicit, if unspecific, nod to over-reliance and addiction, a rare case of a national AI statute naming the companion-app mechanism by description rather than leaving it implicit.

Sources & reading trail

Original Chinese text: promulgation by seven agencies, all 24 articles, and the 15 August 2023 effective date (Article 24).

Source published: 10 July 2023 · Retrieved: 16 September 2026

English-language rendering used to gloss the Chinese original, including Article 10's minor over-reliance duty.

Source published: 13 July 2023 · Retrieved: 16 September 2026

Company documents, filings, studies and official records establish the record; the reading and the questions are Lovebot Journal editorial analysis. This retrospective draft does not imply the site published on the event date.

Continue reading

• A character-chat app quietly runs on a Chinese lab's model
• Companion content filters are a policy choice, not a limit
• Age assurance guidance separates declaring from proving
• Browse the complete the archive

Published September 18, 2026, not on the date of the event described.