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: Chinese AI and Technological Control Becomes Increasingly Sophisticated
In July, China expanded both its international influence over AI governance and its domestic capacity for digital information control. The launch of a new global AI alliance came as Chinese AI models continued to reflect political and ideological restrictions, raising concerns about how these standards may influence AI systems beyond China. At the same time, new surveillance and censorship technologies showed that the Chinese authorities are developing new ways to monitor digital activity and identify tools used to bypass online restrictions.
China’s New AI Alliance and the Global Expansion of Its AI Control Model
On July 17, China launched the WAICO, a 29-country AI cooperation body headquartered in Shanghai, at the annual World Artificial Intelligence Conference. Xi Jinping attended in person for the first time since 2018, describing the organization as a “major move by China to answer the call of the Global South and unite the international community together to vigorously promote AI development and governance.” At the conference, the government also announced that it would provide 5,000 AI training and study opportunities for developing countries over the next five years and expand cooperation with ASEAN, the Arab League, the African Union, and BRICS countries.
Simultaneously, Moonshot AI’s latest Kimi K3 model, launched in July, has reportedly narrowed the performance gap with leading U.S. AI systems, even outperforming OpenAI and Anthropic’s most advanced models on certain benchmarks. Moreover, Deepseek has reportedly begun developing its own AI inference chip. The effort is intended to reduce the company’s reliance on Nvidia and Huawei amid tightening U.S. export controls, while following a broader industry trend toward custom AI chips.
Through these developments, the CCP seeks to shape the global AI landscape and the rules that govern it. China’s growing international role in AI is particularly significant given the extensive political and ideological controls imposed on generative AI within the country. Government censorship does not stop at traditional media or online platforms; political requirements are increasingly incorporated into the design and operation of AI systems themselves. A July 2026 report by the Resilience Innovation Lab and Strategic China Policy Group, found that political requirements have increasingly been embedded into China’s technical standards for generative AI. Within AI safety and security frameworks, content that deviates from Party-approved narratives was treated as a content security risk, resulting in restrictions or removal before the content is even published.
HRIC’s testing of the Kimi K3 model illustrated these restrictions in practice. When asked to evaluate Xi Jinping or the Chinese Communist Party, or to explain what happened in China on June 4, 1989, the model avoided answering and instead suggested changing the topic. By contrast, when asked to evaluate Donald Trump, it provided a direct and detailed reply. These polar opposite responses demonstrate the CCP’s influence over Chinese AI models which are being exported around the globe, hiding information about its own government while freely providing answers about another.
Even without explicit government control, AI systems can become susceptible to the information environments in which they are created and trained. A July study by Meta’s independent Oversight Board tested ten leading AI models, including models from outside China, and found that they refused 34 percent of requests for political criticism involving restrictive countries, compared with 14 percent for more permissive countries. For example, while some models were willing to produce material criticizing leaders in the United States or United Kingdom, they refused similar requests involving China or other restrictive governments.
Taken together, these developments raise concerns that extend beyond China’s growing technological competitiveness. As China produces stronger AI models, develops greater independence in semiconductor production, and expands its role in international AI governance, the political norms embedded in its domestic AI ecosystem may gain wider influence. The question, therefore, is not only whether China can compete with the United States in AI capabilities, but also whether its growing global role could normalize a model of AI governance in which political censorship and ideological conformity are treated as components of AI safety and security.
Growing Sophistication of General Information Control
Beyond generative AI, China’s technical tools for surveillance and information control are also becoming more sophisticated. Researchers from the People’s Public Security University of China developed a forensic technique that can analyze faint electromagnetic signals emitted by smartphones. According to the researchers, the method can identify both which apps are being used and some of the actions being performed on the device, even when a phone is offline, in airplane mode, locked, or encrypted. The technique could give law-enforcement authorities another way to obtain information about digital activity without directly accessing the contents of a phone.
Moreover, the crackdown on VPNs continues. Leaked documents revealed that the ability of China’s “Great Firewall” to identify various mainstream circumvention tools has been significantly upgraded, and that it can identify specific VPNs in use. Thus, authorities may increasingly be able to determine that a user is trying to access uncensored information, but also which type of VPN or circumvention technology is being used to do so. More precise detection would make it easier for the authorities to track which VPNs are in use, identify the services that still work, and decide where to focus future blocking or enforcement efforts.
In Case You Missed It: Other Digital Rights and Cybersecurity News
Great Firewall, Censorship, and Information Control
Censorship, Secrecy, and Speculation Continue After Pilot in Beijing CITIC Tower Collision Identified: After a light aircraft crashed into Beijing’s tallest skyscraper, all mention of the incident—from videos and photos to keywords in online posts—were wiped from popular social media apps in China. Many questions also remain as to the future of aviation regulations following this massive security breach.
Related: After Impact.
Related: HRIC on Twitter/X.
The Party at 105: Adept at Struggle, Even Better at Censorship: The day after Xi Jinping spoke at a ceremony for the 105th anniversary of the founding of the CCP “a video from the Communist Youth League Central Committee, ‘100 Heartwarming Seconds from the Ceremony Celebrating the 105th Anniversary of the Founding of the Communist Party of China,’ was apparently blocked on Tencent’s QQ Space platform for ‘suspected violation of national policies and regulations.’” Commenters reacted sarcastically, pointing to the state’s censorship of its own propaganda videos.
Translation: “How Do These Comments Break the Rules?”: A WeChat blogger’s translated essay describes bewilderment at China’s automated comment-censorship system, which flags seemingly innocuous reader comments as rule violations with no discernible logic.
Minitrue: 718 Leaked Directives from Video Platform LeTV: China Digital Times published 718 leaked censorship directives issued to video platform LeTV between 2017 and 2020, detailing granular content-control demands and reporting requirements from Chinese cyber-authorities.
Translation: Euphemism and Censorship Surround Guangxi Flooding, Dam Breach News: Deadly typhoons and a dam breach at Liulan Reservoir in Guangxi which left over 26 dead triggered heavy censorship. Critical WeChat posts questioning the dam’s failure just a year after a costly safety upgrade were deleted. One surviving essay dissects state media’s euphemistic use of “a gap appeared” instead of “collapsed” or “breached” to describe the disaster.
“Everything Is Sensitive”: Foreign Correspondents’ Club Survey Echoes Netizen Frustration at Unpredictable Red Lines: The Foreign Correspondents’ Club of China’s 2025 survey found journalists facing harassment and access denial on an ever-widening range of topics often unrelated to a story’s actual sensitivity. It echoes domestic netizens’ frustration with unpredictable censorship, where even positive posts get deleted and vague rules leave people guessing.
Translations: On Foreign and Domestic Media and “Man-Made Problems Lurking Behind Natural Disasters”: As China braces for more deadly storms following twin typhoons Bavi and Maysak this month, online news about these tragedies, mainly those which highlight government failings such as inadequate disaster-prevention and disaster-relief efforts as well as inaccurate reporting on the injuries and deaths taking place, have been suspiciously taken down.
HRIC on Twitter/X: In the World Cup third-place match, England defeated France 6-4. This score unexpectedly referenced “64” as in “June Fourth,” China’s top sensitive term which refers to the June Fourth Tiananmen Massacre. A related CCTV News Douyin video garnered 197,000 likes but only 80 comments, showing clear signs of comment control.
Turtles for the Traitors: Based on three years of analysis of WeChat posts from China’s Ministry of State Security, China Digital Times found that the MSS’s definition of intelligence gathering from “external hostile forces” has slowly expanded to cover most areas of everyday life, from dating apps to sea turtles.
AI Propaganda and Social Media Manipulation
The Propaganda Promise of AI: China Daily’s editor-in-chief wrote in Qiushi that AI is turning state media from a “cognitive tool” into an “action tool” capable of directly producing propaganda content for global audiences, though the piece notes such AI-generated content (e.g., a PLA video) has drawn mockery online as “AI slop.”
Cybersecurity & Data Privacy
Navigating Risks From Chinese Apps: A new report details the risk of using Chinese-developed applications, specifically due to their ability to collect and store sensitive personal data which must be turned over to the Chinese authorities on request.
Digital Transnational Repression
HRIC on Twitter/X: A July 2026 report by cyber intelligence firm Graphika revealed that the Chinese government-linked influence operation Spamouflage recently employed a new tactic: distributing altered event leaflets in an attempt to disrupt events critical of the Chinese Communist Party which were held in the United States and Europe.
Civil Society & Resistance Under Pressure
China’s graduate glut: millions enter a job market with little use for them: A record 12.7 million Chinese college graduates are entering an oversaturated job market this year, with AI automation increasingly eroding entry-level positions even as Beijing pushes AI-related job schemes to absorb them.
Technological Competition & State Power
China has AI ‘cyber nuclear weapon’: Chinese company Zhipu AI (Z.ai) released a statement claiming that its AI model is equal to Anthropic’s Mythos model when it comes to finding cybersecurity vulnerabilities.
Facing US export controls, China’s DeepSeek plans to make its own chips: DeepSeek is developing its own AI inference chip to cut reliance on Nvidia and Huawei amid tightening US export controls, joining rivals like OpenAI and Anthropic in pursuing custom silicon.
Apple Intelligence approved for launch in China with Alibaba and Baidu: Apple’s generative AI has been approved for launch in China, with the provision that Apple will integrate Alibaba’s Qwen AI model into Apple’s operating systems and work with Baidu search engine, which already complies with the Chinese government’s censorship requirements.
HRIC on Twitter/X: A Deepseek investor meeting memo shines a light on the long-standing controversy over evading export controls. It quotes founder Liang Wenfeng openly admitting that his company purchases large amounts of restricted and non-compliant chips. Equally noteworthy is that this meeting urged investors not to leak specific numbers or details and prohibited screen recordings or external sharing.
White House accuses China’s Moonshot of stealing Anthropic AI: The director of the U.S. White House Office of Science and Technology Policy has claimed that Moonshot AI “distilled” Anthropic’s Fable to develop its K3 model at a scale that may warrant sanctions.
Recent Incident Highlights Use of Chinese Parts in Biometric Equipment in Police, Coast Guard, and Customs: The Taiwanese government has been using Chinese-manufactured technology for matters which relate to national security. Legislator Lin Chun-hsia suggested that “designated government agencies should not be allowed to decide on purchases of equipment that store biometric or other sensitive data.”