China Information Control & Circumvention History: Policies, Cases, Technologies and Countermeasures

Living research project • expanded public edition: September 18, 2026

This research hub traces the evolution of information control in mainland China through a connected historical model: policy or restriction → named person/case → technical or platform mechanism → documented countermeasure → subsequent state/platform response. It is not simply a list of laws. The goal is to show how legal rules, platform practices, network blocking, identity systems, criminal cases, preservation methods and AI-era controls interact over time.

How to read this page

Five layers are kept analytically separate:

• Law and policy: what official rules say, when they took effect and what authorities state they are intended to accomplish.
• Platform enforcement: account, content, search, hosting, app-store or service-provider actions.
• Network-level interference: DNS manipulation, IP blocking, TLS interference and other technical access controls.
• Identity and data controls: real-name verification, account metadata, provider-held records and identity credentials.
• AI/model behavior: model output, retrieval, moderation, routing and other service layers that black-box users cannot safely conflate.

Official Chinese regulations commonly state goals such as protecting national security, public interests, personal information, market order and a “healthy” online environment. Independent press-freedom and human-rights organizations have documented cases in which these systems also restricted journalism, political expression or access to information. This page records both the official regulatory text and independently documented cases rather than treating either as a complete description of the system.

Historical timeline: policies, people, technologies and responses

1989 — Fax, telephone and early computer networks as information channels

Before public internet access became widespread in mainland China, the 1989 movement already showed a recurring information-control dynamic: new communications technologies opened alternative channels, and authorities responded by monitoring or restricting them. Contemporary reporting described Chinese students overseas sending foreign news reports and photographs into China by fax after June 4. Archival collections also preserve faxes, telephone transcripts and other documents that moved between China and overseas networks during and after the movement.

Contemporary reporting also documented official warnings and monitoring directed at fax machines after the crackdown. This pre-internet episode is important because later cycles—email, websites, social media, VPNs and AI—repeat the same structural contest between distribution, identification, preservation and control.

Sources: TIME — “Fax Against Fictions,” June 1989; UC Berkeley archival finding aid — China Democracy Movement and Tiananmen Incident Archives, 1989–1993.

1998–2000 — Independent websites and the first major online-publisher cases

In 1998, Huang Qi and Zeng Li launched the Tianwang website, initially as a missing-persons service. CPJ documented that it later became a forum for information about abuses of power and politically sensitive topics. Huang was arrested on June 3, 2000; CPJ later described him as the first internet publisher in China prosecuted for his work.

In September 2000, the State Council’s Measures for the Administration of Internet Information Services established a nationwide framework governing internet information services.

Pattern: independent domestic publishing expanded the public information space; enforcement increasingly combined criminal cases with provider and hosting regulation.

Historical information-access responses: overseas hosting, email forwarding, mirror sites, diaspora republication and redundant archiving.

Sources: CAC — Internet Information Services Measures; CPJ — Huang Qi sentencing.

2004–2005 — Shi Tao / 师涛: propaganda directives, email metadata and Yahoo

Shi Tao / 师涛 is a central case in this history. CPJ reported that he emailed notes from a propaganda directive concerning media restrictions to an overseas recipient. The Changsha Intermediate People’s Court sentenced him to ten years in prison on April 27, 2005 for “leaking state secrets abroad.” CPJ reported that the judgment cited account-holder and IP-related information supplied by Yahoo Holdings (Hong Kong) Ltd. as evidence used to identify him.

Why this case matters: it connects traditional propaganda control, cross-border electronic transmission, provider-held identifying data and criminal enforcement. It also shows why encryption of message content and exposure of metadata must be treated as different security questions.

Countermeasure lesson: digital-security practice increasingly considered account identity, metadata, provider jurisdiction, retention and compartmentalization — not only website blocking.

Sources: CPJ — Yahoo data used in Shi Tao prosecution; FreeChina.live canonical Shi Tao profile.

2005–2006 — Zhao Jing / 安替 and platform-hosted speech

CPJ reported that Microsoft removed the MSN Spaces blog of Zhao Jing, also known by the pen name An Ti, on December 30, 2005 after he wrote about the removal of editors at the Beijing News. Microsoft said Chinese authorities had informed the company that the blog violated local law. CPJ reported that the controversy later led Microsoft to modify its approach so that content blocked for users in China could remain accessible elsewhere.

Pattern: multinational platforms were increasingly asked to reconcile local legal requirements with global hosting and speech policies.

Source: CPJ — Zhao Jing / Microsoft case.

2010 — Google ends self-censored search on Google.cn

In March 2010, Google announced that it had stopped censoring Google Search, Google News and Google Images on Google.cn and redirected users to an uncensored simplified-Chinese search service hosted in Hong Kong. Google said the decision followed its January announcement that it was no longer willing to censor search results and came amid disputes over cyberattacks, surveillance concerns and restrictions on online expression.

In June 2010, Google said Chinese officials had informed the company that the automatic redirect was unacceptable and that continuing it would jeopardize renewal of Google’s Internet Content Provider license. Google then replaced the automatic redirect with a Google.cn landing page linking users to Google.com.hk.

Why this matters: the episode is a major example of the tension between local licensing requirements, corporate self-censorship and cross-border provision of search services.

Sources: Google — March 2010 China update; Google — June 2010 China update.

2013 — GitHub blocking demonstrates multiple network-control techniques

GreatFire documented a short but technically revealing period of GitHub interference in January 2013. Its measurements described connection resets, DNS poisoning and later reports of altered SSL certificates. The full-site block was lifted after several days, but GitHub-related paths have continued to experience varying levels of interference over time.

Why this matters: GitHub combines software-development infrastructure with the ability to host documents and censorship-resistance projects. The episode demonstrates why a single label such as “blocked” can hide several different technical mechanisms and why measurement should record DNS, TCP/IP, TLS and application behavior separately.

Sources: GreatFire — GitHub blocking timeline, January 2013; GreatFire — GitHub and reported man-in-the-middle interference.

2015 — Real identity behind public-facing accounts

The 2015 rules on internet user account names formalized the principle commonly summarized as “后台实名、前台自愿”: providers verify real identity behind the account while users may choose a public-facing name, subject to content and identity rules.

Mechanism: a public pseudonym is distinct from anonymity from the service provider.

Historical countermeasure pattern: separation of public-facing identity, private communication, overseas publication and archival roles became more important to privacy-conscious users and organizations.

Source: CAC — 2015 account-name rules.

2015–2017 — WeChat demonstrates account-scoped, silent and dynamic censorship

Citizen Lab’s controlled testing of WeChat documented several mechanisms that cannot be reduced to national network blocking. A 2016 study found that keyword filtering depended in part on whether an account had originally been registered to a mainland China phone number; some blocked messages disappeared silently without notifying either sender or recipient, and group chats showed more censorship than one-to-one chats.

Follow-up research documented censorship of keyword combinations and images relating to the 709 crackdown, Liu Xiaobo and major political events. Citizen Lab emphasized that blocked keyword sets can change in response to events, making censorship dynamic rather than a fixed permanent list.

Research implication: the project should record account origin, conversation type, language, date and exact tested content when evaluating platform censorship. A single successful or failed message is not enough to characterize an entire service.

Sources: Citizen Lab — One App, Two Systems, 2016; Citizen Lab — We (Can’t) Chat, 2017; Citizen Lab — Liu Xiaobo censorship study, 2017.

2017 — Internet-news licensing and cross-border connectivity controls

The 2017 Provisions on the Administration of Internet News Information Services require licenses for internet news information services delivered through websites, apps, forums, blogs, microblogs, public accounts, instant-messaging tools and livestreams. The rules define “news information” to include reporting and commentary on political, economic, military, diplomatic and other public affairs.

Separately, MIIT’s January 2017 campaign to regulate the internet access-service market stated that cross-border telecommunications business could not be conducted through self-built or leased dedicated lines, including VPN channels, without approval.

Mechanisms: licensing, platform management, access-provider controls and regulation of cross-border network services.

Sources: CAC — 2017 Internet News rules; MIIT — 2017 internet access-service market notice.

2015 — Great Cannon: counter-censorship infrastructure becomes an attack target

Citizen Lab analyzed large-scale 2015 distributed denial-of-service attacks against GreatFire.org infrastructure and GreatFire-related GitHub pages. Its researchers described a system they called the Great Cannon, distinct from but co-located with the Great Firewall, that could inject traffic into connections crossing China’s network boundary.

Citizen Lab stated that the observed attacks targeted services designed to make blocked content available in China. The report also cautioned that identifying the precise authority responsible for operating such infrastructure was difficult.

Why this matters: information control can involve not only blocking requests from users inside China but also pressure on infrastructure outside China that helps preserve or relay censored information.

Source: Citizen Lab — China’s Great Cannon, 2015.

2016–2019 — Huang Qi / 黄琦 and 64 Tianwang return as a major case

Huang Qi was detained again in November 2016. In July 2019, the Mianyang Intermediate People’s Court sentenced him to 12 years on charges including “deliberately leaking state secrets” and “illegally providing state secrets to foreign countries,” according to CPJ. His case illustrates that information control is not only technical: it can also involve prosecution of people who collect, verify and publish local human-rights reports.

Source: CPJ — Huang Qi sentenced in 2019.

2020 — Platform governance and the COVID-19 information-control cases

The Provisions on the Governance of the Online Information Content Ecosystem, effective March 1, 2020, require platforms to maintain systems for user registration, account management, information-publication review, comment review, real-time inspection and emergency response. The regulation describes a broad governance model involving government, platforms, content producers and users.

The early COVID-19 outbreak provides a dense cluster of named cases showing the human side of information control:

• Li Wenliang / 李文亮: Human Rights Watch reported that Wuhan police summoned him for “spreading rumors” after he warned colleagues in an online chat about the emerging virus. His later death generated a large online reaction and renewed debate over freedom of expression.
• Chen Qiushi / 陈秋实: CPJ reported that he traveled to Wuhan on January 24, 2020, filmed conditions in hospitals and disappeared on February 6 after saying he planned to report on a temporary hospital.
• Li Zehua / 李泽华: CPJ reported that he livestreamed security agents pursuing him on February 26, disappeared, and reappeared roughly two months later saying he had been placed in supervised quarantine.
• Zhang Zhan / 张展: CPJ documented that she reported from Wuhan through Twitter/X and YouTube, was detained in May 2020 and received a four-year prison sentence in December 2020. After release in May 2024, she was arrested again in August 2024 and sentenced to another four years in September 2025, according to CPJ.
• Chen Mei / 陈玫 and Cai Wei / 蔡伟: Human Rights Watch reported their detention after they worked on archiving censored COVID-19 articles, interviews and personal accounts.

Countermeasure pattern: screenshots, mirrors, GitHub or overseas archives, foreign-hosted video, social reposting, coded language and diaspora preservation became ways to keep deleted material available. These methods do not eliminate legal or security risk.

Sources: CAC — Online Information Content Ecosystem rules; HRW — COVID-19 and expression; CPJ — Chen Qiushi; CPJ — Li Zehua; CPJ — Zhang Zhan.

2022 — Account information, real-name verification and IP-region display

The 2022 Provisions on the Administration of Internet User Account Information require service providers to maintain systems for real-identity authentication, account verification, content security, ecological governance, emergency response and personal-information protection. CAC’s explanatory material states that providers offering information publication or instant messaging should authenticate users’ real identity and display IP-address region information in prescribed contexts.

Mechanism: identity, account metadata and content governance become increasingly integrated.

Source: CAC — 2022 account-information rules.

2022 — Search-engine censorship as a distinct platform layer

Citizen Lab compared eight search platforms operating in China and documented query-based restrictions affecting political and religious information. Its study emphasized that platform-level search censorship differs technically from national network filtering: a search provider can selectively suppress results within an otherwise accessible service, including when HTTPS prevents network infrastructure from inspecting page-level content in the same way.

Research implication: “Can the website be reached?” and “Can the information be found through the platform?” are different measurements and should be recorded separately.

Source: Citizen Lab — Missing Links: A Comparison of Search Censorship in China.

2023 — Generative AI enters the regulatory framework

The 2023 Interim Measures for the Management of Generative Artificial Intelligence Services apply to public-facing generative AI services in mainland China. Article 4 states that providers and users should comply with law and, among other requirements, should not generate content prohibited by laws and regulations involving national security, state power, national unity, terrorism, violence, pornography and related categories. The measures also require attention to privacy, personal information, intellectual property, discrimination and transparency.

New research problem: AI systems can restrict information at the model-output, retrieval, post-generation moderation, routing, account-policy or service layers. Black-box users should not infer which hidden layer caused a refusal without evidence.

Source: CAC — 2023 Generative AI Measures.

2023 — Censorship measurement itself becomes a blocking target

OONI reported that its website and OONI Probe infrastructure became blocked in China beginning July 7, 2023. Its analysis identified DNS injection, TLS interference and IP blocking, and it also found interference with backend services used to submit measurements.

Why it matters: tools used to observe censorship can themselves become targets of censorship.

Source: OONI — China started blocking OONI.

2025 — National network identity credentials

The 2025 Measures for the Administration of the National Network Identity Authentication Public Service, effective July 15, 2025, create a state public-service framework for “网号” and “网证” derived from legal identity information. The regulation says eligible people may voluntarily apply and that the credentials can support non-plaintext identity verification.

Two-sided research question: the system can reduce repeated transmission of plaintext identity information in some scenarios while also further institutionalizing identity verification in online services. Both dimensions should be documented without inferring motives beyond the official text.

Source: CAC — 2025 National Network Identity Authentication Public Service Measures.

2026 — DeepSeek black-box audit and the AI-era evidence problem

Free China Movement’s September 17, 2026 report, From Copying to History: A Black-Box Audit of Politically Sensitive Response Restrictions in DeepSeek, tested controlled prompts involving Tiananmen Mothers, the 1989 Beijing student movement and Lian Shengde.

The audit found behavior consistent with content-sensitive response restriction in the tested interface, including refusals for some mechanically simple operations on supplied sensitive text. Other trials showed that Lian Shengde’s name and supplied 1989-related statements could be reproduced. The evidence therefore does not establish a deterministic single-keyword or single-person blacklist.

Canonical report: DeepSeek Black-Box Audit.

2026 — ZCode cloud-upload controversy and user-control questions

FreeChina.live also documented the ZCode repository-upload controversy. A reverse-engineering report alleged that ZCode could package and upload workspace and Git-history material to Alibaba Cloud OSS; Zhipu AI later acknowledged repository-data uploads tied to its Repo Wiki feature, apologized, and announced remediation. This case is tracked as a data-flow, consent and platform-transparency issue, not as proof of a Great Firewall mechanism or government surveillance. It is relevant to the broader history because information control increasingly depends not only on blocking and deletion but also on platform architecture, cloud routing, identity systems, data access and user visibility.

Dedicated Great Firewall Technical Research

For network-layer censorship mechanisms, measurement datasets and circumvention research, see the dedicated companion page: Great Firewall of China: Technical History, Censorship Mechanisms, Cases, Data and Circumvention. The companion page separates GFW packet/network mechanisms from platform, identity, enforcement and AI-service layers.

Technical control mechanisms: what is actually being measured?

Layer	Examples	What evidence can establish	What it cannot establish by itself
DNS	Injected or falsified DNS responses	Observed name-resolution interference	Who authored a policy decision
IP/network	IP blocking, connection resets	Reachability failure tied to network paths	Content-level model behavior
TLS	TLS handshake interference	Encrypted-session establishment is disrupted	Training-data inclusion/exclusion
Platform	Account suspension, post deletion, search suppression	Observed action by a service/platform	Whether the network layer is also blocking
Identity/data	Real-name checks, subscriber records, IP metadata	Identity linkage or retained account data	Whether message content itself was read
AI	Refusal, deflection, altered retrieval, routing	Behavior of the tested interface/model version	Hidden moderation architecture without internal evidence

Countermeasure history

This section documents circumvention and preservation historically and defensively. It is not a guarantee of safety or legality, and effectiveness changes as blocking techniques change.

Countermeasure family	Purpose	Historical role	Typical response or limitation
Overseas hosting and mirrors	Preserve material outside domestic hosting controls	Dissident publications, archives, rights reporting	Domain/IP blocking, search suppression, access-path blocking
Email/newsletters	Direct distribution	Important before social platforms and in early advocacy networks	Filtering, account monitoring, metadata exposure
HTTPS/encrypted messaging	Protect content in transit	Reduces passive visibility of plaintext content	Endpoint/account controls, metadata, TLS interference
VPNs/proxies	Reach services blocked on the local network	Broad individual and organizational use	Traffic detection, provider blocking, licensing and market controls
Tor bridges / pluggable transports	Make censorship-resistant connections harder to enumerate or classify	Bridges, obfs4, Snowflake, WebTunnel and related systems	Bridge discovery, active probing, transport adaptation
Screenshots / coded language / homophones	Preserve and redistribute removed content	Social-platform environments	Image OCR, semantic moderation, account sanctions
Diaspora archiving / redundant repositories	Preserve evidence after deletion	Historical documents, rights reports, citizen-journalism records	Archive-domain blocking or access restrictions
Censorship measurement	Document blocking methods empirically	OONI and related projects	Measurement sites/backends can themselves be blocked
AI black-box auditing	Measure response differences without internal access	Emerging 2020s method	Requires repetition, controls and version/account/region logging

People, cases and platforms index

Person / case	Period	Significance	FreeChina status
Huang Qi / 黄琦 — Tianwang / 64 Tianwang	1998–2003; 2016–present case history	Early independent website; human-rights publishing; repeated prosecution	Canonical profile LIVE
Shi Tao / 师涛	2004–2013	Propaganda directive, overseas email, Yahoo subscriber/IP data, 10-year sentence	Canonical profile LIVE
Zhao Jing / 赵京 (安替)	2005–2006	Microsoft-hosted blog removal after sensitive media reporting	Canonical profile LIVE
Li Wenliang / 李文亮	2019–2020	Online warning, police admonition, later national debate over information suppression	Canonical profile LIVE
Chen Qiushi / 陈秋实	2020	Independent Wuhan video reporting; disappearance/forced-quarantine reporting	Canonical profile LIVE
Li Zehua / 李泽华	2020	Livestreamed pursuit; disappearance; later account of supervised quarantine	Canonical profile LIVE
Zhang Zhan / 张展	2020–2026	Citizen reporting on Wuhan; two prison sentences documented by CPJ	Canonical profile LIVE
Chen Mei / 陈玫 and Cai Wei / 蔡伟	2020	Archiving deleted COVID-era articles and accounts	Canonical joint case page LIVE
Tiananmen Mothers / 天安门母亲	1989–present; AI test case 2026	Historical-memory advocacy; controlled sensitive-term test case	Canonical profile LIVE
Lian Shengde / 连胜德	1989; AI test case 2026	Historical retrieval versus reproduction controls in DeepSeek audit	Canonical P1 profile LIVE
DeepSeek black-box audit	2026	AI-era controlled response testing	Canonical research report LIVE

Foreign and domestic platform-role matrix

Platform/company	Documented case	Observed role	Evidence caution
Yahoo	Shi Tao	Account-holder/IP-related data cited in court evidence, according to CPJ	Document the specific court-record role; do not generalize to all Yahoo operations
Microsoft	Zhao Jing / 安替	Removed MSN Spaces blog after authorities said it violated local law, according to CPJ	Platform policy changed after criticism; historical context matters
Domestic social platforms	COVID-era and other sensitive topics	Post/account deletion, search moderation and platform governance under domestic rules	Case-by-case evidence is needed to attribute individual removals
DeepSeek	2026 black-box audit	Observed content-sensitive response differences	Black-box behavior does not expose internal architecture

Research data model

Every project row should capture:

Date | Policy/law/directive | Issuing authority | Official stated purpose | Restriction mechanism | Medium/service | Named people/cases | Organization/company/platform | Enforcement action | Technical/data mechanism | Documented countermeasure | Subsequent state/platform response | Primary source | Independent source | P1 profile | Case page | AI/DeepSeek relation | Confidence/evidence status.

Sensitive-information and model-testing protocol

• Do not publish passwords, tokens, live credentials, private contact details or unnecessary unpublished family information.
• Prefer public historical material, controlled strings or redacted examples in AI tests.
• Record exact prompt, model/version, interface/API, date/time, account/region context when relevant, complete response and screenshot.
• Keep training data, retrieval, moderation, routing and final output as separate fields.
• Do not treat a model’s explanation of its own refusal as authoritative documentation of hidden system architecture.
• When a named person appears, the first substantive mention should link to a canonical P1 profile; missing profiles enter the profile-creation queue.
• When a policy or regulation is cited, link to the official text and separately identify independent documentation of effects or contested interpretations.

Version Control & PDF Archive

The web page is the living current edition. Each substantive research release is also preserved as an immutable, citable PDF snapshot.

Version	Date	Status	Major contents	PDF	SHA-256
v2026.09.18.1	September 18, 2026	Current frozen snapshot	1989-to-2026 historical spine; policy/network/platform/identity/AI layer model; named-person cases; platform/company matrix; countermeasure history; research data model; sensitive-information and AI-testing protocol.	Download PDF	a08d885296860d9a88e381b04acb3174360956f827a50fda6fa7411cdc04aa89

Versioning rule: increment the research version for new primary evidence, material factual correction, new major case/policy/technology, methodology or dataset changes, or material countermeasure updates. Preserve prior PDFs rather than overwriting them.

Ongoing Update Lanes

1. Policy/legal chronology: add newly issued CAC, MIIT, public-security, data, algorithm and AI rules with official text, effective date and independently documented effects.
2. Great Firewall measurements: keep the dedicated GFW companion page synchronized with new OONI, GFWatch, Tor, GreatFire and peer-reviewed measurement findings.
3. Named cases and canonical profiles: add newly verified cases, link first substantive mentions to canonical P1 profiles, and maintain status changes without duplicating people.
4. Technology-company matrix: extend provider/platform cases covering data disclosure, search filtering, hosting removal, app-store decisions, cloud/data routing and AI-service behavior.
5. Countermeasure history: document changes in mirrors, VPN/proxy ecosystems, Tor transports, encrypted communications, archiving and censorship measurement, always distinguishing historical evidence from present-day operational advice.
6. AI auditing: repeat preregistered black-box tests across model versions and interfaces while keeping training data, retrieval, moderation, routing and final output analytically separate.

This is a living research page. It will be expanded and corrected as stronger primary evidence becomes available. Historical countermeasures are documented for research, preservation and defensive information-access analysis; effectiveness and legal risk vary by place and time.