Welcome to the HRIC Digital Rights Report, our monthly roundup of developments in China’s digital rights landscape, including cybersecurity, AI, digital transnational repression, censorship, the Great Firewall, and internet freedom.
This Month: AI-Powered Surveillance, PRC-Aligned Misuse of Anthropic’s Claude, and Hong Kong’s “AI City Brain”
Recent developments in China and Hong Kong continue to raise concerns about the use of artificial intelligence for surveillance, intelligence gathering, and monitoring politically “sensitive” individuals and communities. In September 2026, two developments were particularly noteworthy: Anthropic’s disclosure of a suspected Chinese state-linked intelligence operation involving the misuse of its AI models, and Hong Kong’s announcement of plans to develop an AI-powered “City Brain.”
In September, Anthropic published its report Detecting and Countering Misuse of AI, revealing that it had disrupted a suspected China-based, PRC government-aligned intelligence operation involving the misuse of its Claude AI models between December 2025 and August 2026. According to Anthropic, the operators used Claude for intelligence analysis, directing the platform to produce dossiers, including “personnel research drafts,” investigative “clue reports,” and daily “situational awareness” to target religious leaders, diaspora communities, and organizations considered politically sensitive by Chinese authorities.
The targets included senior Catholic cardinals across Asia, the leadership of the Presbyterian Church in Taiwan, the Central Tibetan Administration, Tibetan advocacy organizations such as the International Campaign for Tibet and Students for a Free Tibet, as well as Falun Gong practitioners and affiliated media organizations. Anthropic also reported that China-based actors linked to public and state security organs used Claude to target organizers of Tiananmen Square commemorations.
The operators collected extensive personal information, including individuals’ dates and places of birth, immigration dates, and social media accounts. They also conducted reconnaissance of religious facilities, gathering floor plans, photographs of building facades, and structural diagrams. Their monitoring extended across Chinese and international social media platforms, including WeChat, Xiaohongshu, Douyin, Weibo, LinkedIn, Instagram, Threads, X, and Facebook.
These findings illustrate how generative AI can reduce the time and resources needed to compile profiles and analyze information about surveillance targets. (There is no evidence that these operators accessed other users’ private Claude conversations.) This is only one of several recent examples in which state agents used various AI tools to develop sophisticated cyberattacks and intrusions, obtain confidential information, or simply manipulate AI-generated content. In an earlier disclosure, OpenAI reported that a person associated with Chinese law enforcement had used ChatGPT to edit reports concerning operations against dissidents and foreign political figures.
Anthropic’s report underscores the risks of AI tools and the responsibility of AI developers to detect and disrupt misuse. Platforms should exercise diligence when handling user information and promptly inform the public of any identified interference or misuse. Human rights defenders should also exercise caution when sharing personal or confidential information with AI tools and platforms.
On September 16, Hong Kong Chief Executive John Lee announced plans to develop an “AI City Brain” in his annual Policy Address. The proposed system would use AI to integrate and analyze data from Hong Kong’s government departments and digital infrastructure in real time, with the stated aim of improving urban management and emergency response. Although a senior official subsequently said the system was not intended to surveil individuals, the proposal notably includes law-enforcement tools such as the Security Bureau’s Emergency Monitoring and Support Centre and a police-led Inter-departmental Major Incident Platform. Their integration raises concerns about how the system could be used to monitor individuals, particularly without clear limits on data access, retention, and independent oversight.
The risks associated with the AI City Brain extend beyond conventional surveillance. Integrating large amounts of citizens’ personal data into an AI-powered system introduces additional concerns related to data breaches, unauthorized access, and misuse. Such integration would also significantly expand the authorities’ ability to comprehensively monitor individuals, connect information from different sources, and infer patterns of behavior. AI-generated assessments may also contain errors or biases that could harm individuals if used in official decisions. These risks warrant particular scrutiny in Hong Kong’s increasingly restrictive political environment.
In Case You Missed It: Other Digital Rights and Cybersecurity News
Great Firewall, Censorship, and Information Control
CAC expands campaign against AI-generated content: On September 2, the Cyberspace Administration of China reported cumulative removal of more than 5.61 million items, action against over 49,000 accounts, and measures against more than 2,400 websites and apps in its AI campaign. These are official cumulative figures, not September-only totals. The notice describes expanded face and voice databases and tighter screening of training material and AI outputs, linking anti-abuse enforcement with broader content controls.
Two-month crackdown on fabricated short-video personas: On September 30, the CAC announced a campaign targeting staged hardship narratives, false professional identities, AI impersonation, and manufactured conflict. While the stated targets include fraud and deceptive marketing, broad enforcement could also potentially chill legitimate accounts of poverty and social grievances.
As the Chinese Communist Party (CCP) covers up the extent of the mudslide disaster in Tibet, several netizens have been arrested for posting related information: Chinese authorities have censored footage and punished at least ten citizens for “spreading rumors” online about death tolls following the catastrophic August 26 Gyirong Port mudslide on the Tibet-Nepal border, while barring foreign journalists from accessing the disaster zone.
Minitrue: Find, Delete, and Block Chai Jing Video on Cultural Revolution Leader: A leaked CAC directive ordered platforms to find, delete, and block a Chai Jing documentary segment on Gang of Four figure Wang Hongwen using video fingerprinting technology, part of broader efforts to suppress Cultural Revolution history.
Translation: Article on Media’s Survival Fails to Survive: A frank WeChat article from Zhejiang’s provincial propaganda department candidly admitted the structural crisis facing Chinese local media, including control by tech platforms and shrinking revenue, was censored within hours, illustrating how even officially sanctioned honesty about media decline is unwelcome.
Minitrue: ByteDance Censors’ List of 5,241 Sensitive Terms: CDT published and analyzed a leaked 2020 internal ByteDance moderation lexicon of 5,241 sensitive words and phrases covering political figures, historical events, and locations, used to censor content across TikTok’s sister apps Douyin and Toutiao.
Translation: “Eventually, We All Learned to Keep Quiet”: Prompted by censorship following a Guangzhou mass stabbing which targeted women, this essay reflects on how self-censorship has become an internalized survival skill in China, warning that suppressed anger curdles into dangerous societal numbness rather than genuine social harmony.
Scripting the Source: Chinese state TV journalists reveal that live interviews are increasingly pre-scripted before filming even begins, with interviewees given talking points to recite. This “risk management” shift strips journalism of its function as authentic, on-the-spot information gathering.
China’s Online Environment Gets Colder and Darker: Li Yuan describes persistent pessimism on Chinese social media amid economic slowdown and unemployment, challenging the official “positive energy” narrative even as space for substantive criticism narrows.
AI Propaganda and Social Media Manipulation
HRIC on X: People’s Daily opened an account on Threads, complete with a blue verification checkmark, posting feel-good content like Guangzhou night scenes, dragon boat races, and light shows. One post garnered over three thousand views and just eleven likes, but the replies painted a starkly different picture: one comment paired an old photo of Xi Jinping sitting at a computer with the caption “Watch me cut off your VPN.”
HRIC on X: Huanyu Zhiying 1.0, unveiled by Sun Yat-sen University in April, can simulate foreign-media coverage, predict how different audiences might react, then refine Chinese messaging in response. Instead of broadcasting a narrative and seeing what lands, researchers can potentially test it against synthetic audiences first.
HRIC on X: ASPI’s analysis of the “Global Audience Intelligence Response 1.0” system describes more than 10,000 simulated personas covering 18 countries. The system uses survey and social media data to model foreign audience reactions and refine messaging, raising concerns about AI-assisted influence operations.
Digital Policy and Regulation
New rules for multi-channel networks take effect September 1: The rules require registration, identity verification, content management, and platform reporting for organizations supporting online creators. They also require adherence to official political and public-opinion priorities and prohibit broadly defined conduct involving negative information and divisive topics. These provisions raise concerns about further restrictions on online expression.
China’s top court posts guidelines on deepfakes, voice cloning, and algorithmic price discrimination: The Supreme People’s Court issued guidance addressing unauthorized deepfakes and voice cloning, platforms’ responsibility for responding to infringing content, algorithmic price discrimination, and AI-generated misinformation.
Policy Address 2026: Hong Kong to appoint AI commissioner and consider deepfake laws – leader John Lee: In his Policy Address, John Lee announced a new AI Commissioner post to combat AI-enabled crime and said the government will study deepfake legislation, following cases of scammers using cloned voices to steal over HK$10 million.
Beijing bans drone possession in the city after earlier moves to tighten rules: Beijing will ban possessing drones within city limits starting mid-November, extending an earlier May sales ban, following a fatal June crash in which a small aircraft struck a Beijing skyscraper. The authorities will offer buyback subsidies for existing owners.
Cybersecurity & Data Privacy
The Global Data Center Boom Is a Gift to Spies: shared “colocation” data centers, where Chinese and American companies often lease space in the same facility, create espionage risks far simpler than hacking, as tenants can physically tamper with neighboring servers.
HRIC on X: According to the New York Times, California cybersecurity firm Calif recently disclosed a high-risk “zero-click” worm attack targeting WeChat, dubbed “WeWorm.” Victims do not even need to click links, download files, or answer calls—a single WeChat call from a compromised contact could serve as the entry point for intrusion.
HRIC on X: Restrictions on direct access to leading foreign AI services such as ChatGPT and Claude have encouraged corporate employees in mainland China to use third-party AI relay services. Requests for help with coding, system maintenance, and troubleshooting may expose internal code, server logs, VPN configurations, and authentication credentials to those intermediaries.
Digital Transnational Repression
HRIC on X: A new phishing campaign is targeting Chinese human rights defenders and activists with fake party invitations that appear to come from familiar contacts and trusted invitation services. These messages may lead to fake login pages designed to steal passwords, authentication codes, or active sessions.
Chinese Online Harassment and Death Threats Against Me Continue as I Speak for Uyghurs: This personal essay is part of an ongoing series in which the Uyghur Times founder recounts continued online death threats and hacking attempts linked to Chinese state networks, alongside the detention of dozens of his relatives in China in apparent retaliation for his journalism abroad.
Technological Competition & State Power
Chinese AI firms accused of unauthorized model distillation: US agencies accused six Chinese AI developers of extracting capabilities from leading US models through fraudulent accounts and access-evasion methods; China rejected the allegations. Separately, Anthropic reported blocking distillation campaigns linked to seven Chinese labs. The allegations concern different sets of actors and should not be treated as a single finding.
HRIC on X: Chinese judicial materials describe criminal prosecutions over unauthorized cross-border transport of Starlink terminals, highlighting restrictions on access to satellite communications outside state-controlled networks.
China’s AI security debate: Chinese officials rejected calls to constrain China’s AI development while intelligence chief Chen Yixin warned about AI-enabled disinformation and cyberattacks. The debate illustrates how authorities frame AI as both a strategic asset and a threat to political security.
Technology controls and freedom of movement: New Chinese rules permitting exit bans for technology-export violations prompted Taiwan to warn technology workers of detention risks, connecting technology competition with restrictions on individual mobility.
US–China AI talks: The two governments agreed to establish a communication channel for AI-related incidents and hold an AI dialogue in November. The practical implications for rights protections and accountability remain unclear.
Open-Weight Diplomacy: This analysis argues that China’s low-cost, open-weight AI models could deepen overseas dependence on Chinese chips, cloud infrastructure, and training services, extending the Digital Silk Road model into AI.
Current Events
HRIC on X: China’s youth employment landscape is already dire, and AI is exacerbating the problem. A record-breaking 12.7 million graduates are entering the job market this year, and the unemployment rate among urban youth aged 16 to 24 rose to 17.9% in July.