China

Article 14 of the Interim Measures for the Management of Generative AI Services requires a provider that discovers illegal content to promptly stop generating it, stop transmitting it, eliminate it, carry out rectification such as retraining the model, and report to the relevant competent authority.

The article separately requires a provider that discovers a user has used the service for illegal activity to warn, restrict, suspend, or terminate that user's access, preserve relevant records, and report to the relevant competent authority, and it states no promptness standard for that second reporting duty.

Both duties name only the relevant competent authority as addressee, without naming the Cyberspace Administration of China by title, and this reporting duty is distinct from the content-labeling duty the same Act imposes at Article 12.

Article 10 of the Provisions on the Administration of Deep Synthesis Internet Information Services requires a deep synthesis service provider to strengthen its management of deep synthesis content, review users' input data and synthesis output by technical or manual means, build a database of features for identifying illegal and undesirable information, and record and retain the related network logs.

Where the provider discovers illegal or undesirable information, the article requires it to take disposal measures and preserve relevant records in accordance with law, and to promptly report the discovery to the cyberspace administration department and the relevant competent authority, while separately taking measures such as warning, restricting features, suspending service, or closing the account against the user responsible.

Only the reporting step in that sentence carries the word 'promptly' in the text; the disposal and record-preservation steps that precede it carry no separate promptness standard of their own.

Providers of generative AI services to the Chinese public must label generated content such as images and video in accordance with the Deep Synthesis Provisions' Article 16 and 17 labeling rules. This is a cross-reference duty binding generative-AI providers specifically to the general deep-synthesis labeling regime, not a freestanding labeling standard of its own.

Article 7 of the same instrument governs training-data lawfulness and has no labeling content; it is training-data territory outside this topic's scope, correcting an earlier derivation that had cited it here.

Internet information service providers must add an explicit, user-perceptible marker to AI-generated or AI-synthesized text, image, audio, video, and virtual-scene content, and must embed an implicit, machine-readable identifier in file metadata carrying generation attributes and the provider's name or code. Content-distribution platforms must check incoming content for labels and add a risk prompt to unlabeled or suspected-generated content.

No person or organization may maliciously delete, alter, forge, or conceal a required label. The accompanying mandatory national standard GB 45438-2025 sets the technical parameters for both label types and takes effect the same date; no official standards-body URL for GB 45438-2025 is available independent of a commercial reseller, so it is described here rather than authored as its own instrument.

Deep-synthesis service providers must apply a technical measure to label content generated or edited using their service without interfering with normal use, and retain log information as required.

Where a listed category of service could cause public confusion, simulated dialogue or writing, synthesized or altered voice, face generation or face-swap, or immersive simulated-scene generation, the provider must apply a prominent label alerting the public to the deep-synthesis nature of the content. This is the earliest of China's four labeling instruments and is the one the 2023 Generative AI Interim Measures cross-references.

China's first dedicated facial-recognition regulation requires voluntary, explicit, separately-obtained consent before capturing facial information, with guardian consent for a minor under 14, and requires using the method with the least impact on individual rights available.

It mandates that facial data be stored on the collecting device rather than transmitted over the internet absent a legal exception or separate consent, caps retention at the minimum necessary period, bars facial recognition as the sole identity-verification method wherever another method exists, bans facial recognition devices inside private spaces within public venues such as hotel rooms and changing rooms, and requires a processor holding facial data on 100,000 or more people to file with the provincial cyberspace authority.

No comparably dedicated voiceprint regulation exists; a January 2026 CAC draft for public comment would extend this same on-device-storage approach to fingerprint and voiceprint data, but it had not been finalized as of the date shown and is not authored as its own instrument here. Article 2 carves facial-recognition R&D and algorithm-training activities out of the Measures' scope, so the duties below bind deployment and use, not model training (第二条: 不适用本办法的规定).

Upon discovering an actual or possible leak, tampering, or loss of personal information, a handler must immediately take remedial measures and notify both the department responsible for personal information protection and the affected individuals, with the notice covering the categories of information involved, the cause, possible harm, remedial measures, mitigation steps for individuals, and the handler's contact information.

Individual notice may be omitted where remedial measures can be shown to effectively prevent harm, unless the supervisory department requires notice anyway. The statute requires action taken immediately rather than setting a fixed numeric deadline such as 72 hours.

China's omnibus personal-data statute (Ge Ren Xin Xi Bao Hu Fa, Personal Information Protection Law) requires a lawful basis, most commonly informed consent, before processing personal information, and otherwise contract necessity, statutory duty, public-interest journalism, information the individual or another lawfully disclosed, or another legal ground.

It imposes purpose-limitation and data-minimization duties, requires a handler that entrusts processing to a third party to supervise that processor's activity under an agreement, and assigns joint and several liability among two or more processors who jointly decide a shared processing purpose and method. Chapter II, Section 3 (Arts. 33-37) extends parallel duties to state organs, so the Act binds government processors as well as private ones.

Article 38 requires one of a CAC-organized security assessment, personal-information-protection certification, a CAC standard contract, or another state-recognized mechanism before transferring personal information outside China, plus assurance the overseas recipient meets PIPL's protection standard. Article 39 requires advance notice naming the recipient and purpose and separate consent.

Article 40 requires domestic storage for critical information infrastructure operators and above-threshold handlers, with export gated on a security assessment. The 2024 CAC Provisions on Promoting and Regulating Cross-Border Data Flows (a separate instrument) substantially relax these mechanisms by transfer volume.

This CAC regulation relaxes PIPL Article 38's cross-border transfer mechanisms by transfer volume, recited in the CAC promulgation notice. Exporting under 100,000 individuals' non-sensitive personal information in a year, or a listed necessary-business category such as contract performance, cross-border human-resources management, or emergency protection of life or property, needs no security assessment, standard contract, or certification at all.

Exporting 100,000 to under 1,000,000 individuals' non-sensitive personal information, or under 10,000 individuals' sensitive personal information, needs a standard contract or certification. Exporting 1,000,000 or more individuals' non-sensitive personal information, or 10,000 or more individuals' sensitive personal information, or any export by a critical information infrastructure operator, needs the full CAC security assessment. Free trade zones may adopt their own negative lists further narrowing what requires a mechanism.

Individuals have the right to know about and decide on processing, and to limit or refuse it (Art. 44), the right to access and copy their data including portability to another handler under state conditions (Art. 45), the right to correction or completion (Art. 46), the right to deletion in enumerated circumstances with cessation of active processing as a fallback where deletion is technically infeasible (Art. 47), and the right to an explanation of processing rules (Art. 48).

Article 50 requires a convenient exercise mechanism, a reasoned explanation for any refusal, and permits a court suit if a request is unreasonably denied.

Article 24 separately requires transparent, non-discriminatory treatment of automated-decision-making results, prohibits unreasonable differential pricing, requires a non-personalized option or opt-out where automated decisions push marketing, and gives an individual a right to explanation and to refuse a decision made solely by automated means where it has a major effect on their rights, with Article 55(2) requiring a personal-information-protection impact assessment before deploying automated decision-making.

The Cyberspace Administration of China leads coordination, with sectoral State Council departments and their local counterparts supervising within their own scope; investigative powers include interviews, record examination, on-site inspection, and, with approval, seizure of equipment.

Ordinary violations draw a warning, confiscation of gains, and service suspension, escalating on non-correction to fines up to RMB 1,000,000 for the entity and RMB 10,000 to 100,000 for the responsible individual; grave violations draw fines up to RMB 50,000,000 or 5 percent of the prior year's revenue, possible business suspension or license revocation, individual fines of RMB 100,000 to 1,000,000, and a bar on serving as a director, supervisor, senior manager, or protection officer of a related enterprise.

A handler that cannot prove itself free of fault bears civil compensation liability, and procuratorates and designated organizations may bring public-interest suits over mass violations.

Article 28 defines sensitive personal information as information whose leak or misuse would easily harm dignity or personal or property safety, expressly including biometric identification information (covering both faceprint and voiceprint as instances of the same category) alongside religious belief, specific identity, medical health, financial accounts, and location tracking, and extends automatically to a minor's information under 14.

Article 29 requires separate, explicit consent before processing any sensitive personal information, layered on top of the general lawful-basis requirement, Article 30 requires advance notice of necessity and impact, and Article 31 requires guardian consent for a minor's data. The biometric definition carries no exclusion for an identifier extracted from an existing photo, video, or audio recording.

Article 285(1) criminalizes intrusion into state-affairs, defense, or advanced-science computer systems; Art. 285(2), added by the 2009 Amendment VII, reaches intrusion into other systems, or other technical means, to obtain their data or illegally control them, where the circumstances are serious; Art. 286 criminalizes destroying a computer system's function, data, or programs.

These target circumvention and intrusion, not mere collection of what a system already makes public, and no top-level case construing authorization as narrowly as the United States' Van Buren was found for a scraping fact pattern.

This instrument is secondary-sourced only: official primary text was sought at eight URLs across five official domains (npc.gov.cn, flk.npc.gov.cn, spp.gov.cn, court.gov.cn, mps.gov.cn), and every one serves either a JavaScript shell, a server error, or (at flk.npc.gov.cn) a single-page app whose article text loads by a client-side mechanism; no CAPTCHA appears anywhere.

This is a confirmed access-tier gap, not a finding about the law, and it must not be presented as verified against primary text.

Article 24 sets out a closed list of permitted uses that do not require the copyright holder's consent or payment: personal study, brief quotation for commentary, unavoidable reproduction in news reporting, and reprinting current-affairs commentary absent a reservation, among others. There is no general fair-use clause and no text-and-data-mining exception, so training an AI model on scraped copyrighted content has no dedicated statutory basis under this article.

Courts have so far worked the fair-use-adjacent question through AI output cases rather than collection: the Guangzhou Internet Court found AI-generated images infringing in February 2024 (secondary-sourced, the Ultraman case), while the Hangzhou Internet Court held in September 2024, upheld that December, that training itself can be lawful absent intent to plagiarize expression or unreasonable prejudice to the rightholder (secondary-sourced); neither directly answers whether scraping for training is itself permitted.

Article 27 permits a personal information handler to process, within a reasonable scope, personal information an individual has disclosed themselves or that is otherwise lawfully public, unless the individual has expressly declined; processing that has a significant impact on the individual still needs separate consent.

Article 13(6) states the same ground as a general lawful basis for processing, functioning as China's analog to General Data Protection Regulation (GDPR) Art. 6(1)(f), with its own reasonable-scope ceiling and opt-out rather than a legitimate-interest balancing test.

The 2023 Generative AI Interim Measures additionally require a lawful source for AI training data, which for personal information means satisfying this Article 27 basis; that measure's own robots.txt-related technical standard is not independently confirmed and is not recorded as its own instrument here.

The third paragraph of the amended Article 13 prohibits a business operator from obtaining or using, by fraud, coercion, or by evading or destroying technical management measures, data lawfully held by another business operator, where doing so harms that operator's lawful rights and interests and disrupts market competition order. This is China's first purpose-built statutory scraping clause and took effect October 15, 2025, with no reported case applying it yet.

It codifies the result courts had already reached under the law's older general clause: Beijing IP Court held in Sina Weibo v. Maimai (Dec.

2016, secondary-sourced) that reusing API data after a partnership ended was unfair competition under a triple-authorization principle, and Shanghai Pudong New Area People's Court held in Hantao (Dianping) v. Baidu (26 May 2016, secondary-sourced) that spidering reviews into a competing product was an unfair substantive substitute for the source's own product.

A provider of a network product or service must ensure the product or service meets the mandatory requirements of the relevant national standard and must not embed a malicious program in it. On discovering that its product or service has a security defect, vulnerability, or other risk, the provider must immediately take remedial measures and, as provided, promptly notify users and report to the competent authority.

The provider must continuously maintain the product's or service's security and must not terminate that maintenance within the period fixed by regulation or agreed with the user. Where the product or service collects user information, the provider must disclose that collection and obtain consent, and, where personal information is involved, comply with this Law's and the Personal Information Protection Law's rules on personal information.

An ordinary violation draws a corrective order, a warning, and a fine of 50,000 to 500,000 yuan if the provider refuses to correct or the violation endangers network security, with a further escalation, cross-referenced from Article 61's third paragraph, of 500,000 to 2,000,000 yuan for causing a large-scale data leak or the loss of a local function of critical information infrastructure, and 2,000,000 to 10,000,000 yuan where critical information infrastructure loses a primary function.

Any organization or individual engaging in a data-processing activity must, under the law and administrative regulations, establish and improve a full-process data-security management system, organize data-security education and training, and take the technical measures and other necessary measures appropriate to keep the data secure.

Where a data-processing activity is conducted using the internet or another information network, this data-security duty applies on top of, not instead of, the network operator's classified-protection duty under the Cybersecurity Law. A processor of important data must designate a person responsible for data security and a managing body, and implement that responsibility.

An ordinary violation draws a corrective order and a warning, escalating to a fine of 50,000 to 500,000 yuan, with a further escalation to 500,000 to 2,000,000 yuan for refusing to correct or causing a large-scale data leak or another serious consequence; a violation of the separate national core data management system carries its own, heavier fine of 2,000,000 to 10,000,000 yuan and can draw criminal liability, but that heavier tier is not this duty.

Any organization or individual engaging in a data-processing activity must strengthen risk monitoring and, on discovering a data-security defect, vulnerability, or other risk, immediately take remedial measures. On an actual data-security incident, the organization or individual must immediately take disposal measures, promptly notify affected users as provided, and report the incident to the competent authority.

The statute sets no numeric deadline for that notice or report, only immediacy and promptness; the National Cybersecurity Incident Reporting Measures set an operative reporting clock for such an incident, but bind by a network-operator classification recorded in the jurisdiction summary rather than raised as an instrument here.

The penalties for a violation are the same tiered fines as this jurisdiction's Data Security Law data-security-program row, since Article 45 punishes a failure under Articles 27, 29, and 30 together.

A network operator that builds or operates a network, or provides a service through a network, within the territory of China must report a cybersecurity incident under these Measures. These Measures define a network operator as the owner, manager, or network service provider of a network.

These Measures also define a cybersecurity incident as an event that, due to human causes, a network attack, a network vulnerability or hidden danger, a hardware or software defect or malfunction, force majeure, or another factor, harms a network or information system or the data and business applications within it and has a negative impact on the state, society, or the economy.

On discovering or becoming aware of an incident involving its own unit, a network operator must grade the incident against the annexed National Cybersecurity Incident Classification and Grading Guide and report only where the grade is relatively major or above. Where the incident involves critical information infrastructure, the operator must report to the sector's protection-work department and to the public security organ within 1 hour of discovering or becoming aware of it.

A network operator that is a department of a central or state organ, or a directly affiliated unit of one, must report to its own department's cyberspace affairs unit within 2 hours. Every other network operator must report to the cyberspace administration department of its own province within 4 hours.

A major or especially major incident escalates further between the government bodies that received the operator's report, up to the national cyberspace administration, without imposing any additional step on the reporting operator itself.

A network operator must require, by contract, any organization or individual providing it network-security or system-operation-and-maintenance services to promptly report to it any cybersecurity incident that provider discovers through monitoring, and to assist the operator's own reporting under these Measures.

Within 30 days of completing disposal of a relatively major incident or above, the operator must submit a summary report on the incident's cause, its emergency response, the harm caused, accountability, remediation, and lessons learned, through the same channel used for the original report.

A failure to report as required draws a penalty under other law, and reporting late, missing a report, or filing a false or concealed report that causes major harm draws a heavier penalty on the operator and its responsible persons. An operator that took reasonable and necessary protective measures, disposed of the incident under its emergency plan, effectively reduced the incident's impact, and reported it as required may be treated leniently or not held accountable at all.

A report involving a state secret follows the rules of the relevant department instead of the general channel described here.

Limits all online game service providers to offering minors no more than one hour of game time per day, only between 20:00 and 21:00 on Fridays, Saturdays, Sundays and legal holidays, with no game service to minors at any other time. Requires every online game account to be registered under a real identity linked to the National Press and Publication Administration's national anti-addiction real-name verification system, and bans guest-mode access.

Requires providers of online games, livestreaming, audio-video and social networking services to build in time management, permission management and spending management functions for minor users, and requires online game providers to use a unified national electronic identity system to verify that game accounts are registered under the user's real identity, with no game service offered to minors between 22:00 and 08:00.

Implementing regulation requiring game, livestream, audio-video and social platforms to provide a minor mode limiting usage time, functions and content per national standards, to reasonably cap single and cumulative spending by minors by age band, to prevent traffic-driven inducements such as fan voting campaigns aimed at minors, and to display age-appropriate ratings prominently at download and login.

Bars livestreaming service providers from issuing a livestreaming anchor (content creator) account to anyone under 16, and requires providers to obtain a parent or guardian's consent and verify identity before issuing an anchor account to a 16 or 17 year old.

Article 23 gives a network service provider that offers a search or link service a safe harbour from damages liability once, on receiving a rights holder's written notice, it disconnects the link to the allegedly infringing work, unless it knew or should have known that the linked work was infringing, in which case it bears joint infringement liability.

That safe harbour runs through the same notice-and-counter-notice mechanism the Regulation gives a hosting provider under Article 22: Article 14 requires a rights holder's takedown notice to name the rights holder, identify the work and its network address, and include preliminary proof of infringement; Article 15 requires the provider to act on it immediately, forwarding the notice to the party that supplied the material; and Articles 16-17 let that party submit a written counter-statement to have the material or link restored, after which the rights holder may not issue a further notice over the same material.

The Regulation was promulgated by State Council Order No. 468 on 18 May 2006 and amended by a State Council decision dated 30 January 2013; the currently effective text's own commencement article states an effective date of 1 July 2006. No reported Chinese decision applies Article 23 specifically to a news aggregator's display of links to third-party journalism, as opposed to file-hosting or search-engine link cases generally.

Article 5(2) excludes 单纯事实消息 (unofficial rendering: "the news of the day", i.e. pure factual news reports) from copyright protection outright, so a bare news item is never a protected work under Chinese law whichever outlet reports it first.

Article 24 separately sets a closed list of uses that need neither the copyright holder's consent nor payment provided the author and title are credited and the use does not unreasonably prejudice the copyright holder: item (2) permits appropriately quoting a published work in another work to introduce, comment on, or explain it; item (3) permits unavoidably reproducing or quoting an already-published work when reporting news in a newspaper, periodical, radio, or television broadcast; and item (4) permits a newspaper, periodical, radio station, or television station to reprint or broadcast another such outlet's already-published current-affairs article on political, economic, or religious questions, unless the copyright holder has declared reprinting or broadcasting impermissible.

None of these exceptions is capped at a headline-length or short-extract threshold, and no reported Chinese court decision applies them to a systematic news aggregator's reproduction of headlines and snippets as opposed to conventional news reporting or quotation.

Article 5 requires any organisation providing an internet news information service to the public, including a news-reprinting service or a news-dissemination platform service, to first obtain an internet news information service licence, and bans operating such a service without one or beyond its scope.

Article 6 conditions that licence on the applicant being a legal person lawfully established within China, having a PRC-citizen principal officer and editor-in-chief, and maintaining dedicated editorial, content-review, and technical-security staff and systems; an entity applying specifically to gather and publish news must itself be a news organisation or a unit under a news-and-propaganda department's supervision.

Article 11 requires a licensed provider to appoint an editor-in-chief who bears overall responsibility for its news content and to register that person with the cyberspace administration.

Article 15 conditions any reprinting of news information on sourcing it only from central news organisations or provincial-level and above news organisations within a state-designated list, on crediting the original source, author, original headline, and editor, on not distorting the original headline's meaning or the news content, on keeping the source traceable, and on separately complying with copyright law.

This is a licensing and content-control regime enforced by the Cyberspace Administration of China, distinct from and additional to the Copyright Law's own exceptions, and no equivalent duty runs to an aggregator that only links to, rather than reprints, another outlet's news.