China’s AI Strategy, Digital Humans, and Governance

China’s governance apparatus has long drawn on an integrated architecture of surveillance, social management, and ideological control, and the emergence of artificial intelligence has deepened and extended each of these dimensions. The country’s approach to AI is inseparable from its broader political system, a system the Chinese Communist Party itself characterizes not as communism but as “socialism with Chinese characteristics” (中国特色社会主义). The distinction matters doctrinally. Deng Xiaoping first articulated the formula at the 12th National Congress of the CCP in September 1982, and the party’s own constitutional text holds that China remains in the “primary stage of socialism” (社会主义初级阶段), a stage estimated to last at least one hundred years from 1949. Communism, in the CCP’s framework, is the party’s ultimate ideal, not a description of the present order. The 2018 amendment to the PRC Constitution reinforced this architecture by adding to Article 1 the statement that “leadership by the Communist Party of China is the defining feature of socialism with Chinese characteristics,” the first time CCP leadership had been stated in the main body of the constitution rather than only in its preamble. The same amendment removed presidential term limits. Xi Jinping Thought on Socialism with Chinese Characteristics for a New Era was simultaneously enshrined in the constitutional preamble, making Xi the first leader to have his guiding ideology constitutionally codified during his own tenure. This is the political system within which China’s AI ambitions and digital human industry have taken shape.

The surveillance infrastructure that underpins much of China’s AI ecosystem rests on two complementary systems. Skynet (天网), first deployed in 2005 as a national video surveillance program originating from pilot work in Chengdu, had grown to encompass over twenty million cameras by 2017 and approximately two hundred million by 2019, with estimates reaching roughly six hundred million by 2025. Eight Chinese cities rank among the world’s ten most surveilled. Skynet’s capabilities include real-time facial recognition, license plate recognition, gait recognition, and predictive policing, powered by companies including Megvii, Hikvision, Dahua, SenseTime, and Huawei. Sharp Eyes (雪亮工程), authorized nationally in 2015 under a joint directive from the National Development and Reform Commission and eight other ministries, extends this surveillance architecture into rural and semi-urban areas through a county-township-village three-tier system that complements Skynet’s urban coverage. The system’s name derives from a Mao-era phrase, “the people have sharp eyes,” and it was included in the 13th Five-Year Plan with a goal of full coverage of public spaces by 2020. As of 2023–2024, Nanchang police had procured over four thousand new intelligent cameras under Sharp Eyes, and Hong Kong Police began integration with the system in April 2025, assisting in 292 criminal cases and 512 arrests.

Alongside these visual surveillance networks, China’s Social Credit System (社会信用体系), authorized by the State Council in June 2014, operates across four pillars: government integrity, commercial integrity, social integrity, and judicial credibility. A widespread misunderstanding frames the system as a unified national scoring mechanism for individuals. In practice it remains a fragmented collection of different systems, and most local individual scoring pilots have ended as of 2025. The system’s operational center of gravity has shifted primarily to corporate social credit. The National Credit Information Sharing Platform holds over 80.7 billion records covering approximately 180 million businesses, and a Social Credit Construction Law remains in draft as of April 2026. In Xinjiang, the Integrated Joint Operations Platform, introduced in August 2016, represents the most intensive application of AI-driven predictive policing, aggregating data from CCTV, Wi-Fi sniffers, checkpoint scanners, smartphone monitoring, vehicle tracking, and utility records to flag behaviors deemed suspicious, including VPN use, WhatsApp usage, and unusual travel patterns. Human Rights Watch documented the system through reverse engineering of its mobile application in 2019. Huawei cooperated with Megvii in 2018 to test a camera system capable of identifying ethnicity and triggering what was described as a “Uyghur alarm.”

The strategic framework animating China’s AI development across all these domains was codified on July 8, 2017, when the State Council issued the New Generation Artificial Intelligence Development Plan (新一代人工智能发展规划), document number 国发〔2017〕35号. The plan established a three-step timeline. By 2020, China’s AI technology and applications were to keep pace with the world’s advanced level, with the core industry reaching over 150 billion yuan. By 2025, some technologies and applications were to reach world-leading level, with the core industry exceeding 400 billion yuan. By 2030, AI theory, technology, and applications were to reach world-leading level overall, with China becoming “a major world AI innovation center” (世界主要人工智能创新中心) and the core industry surpassing one trillion yuan. The phrasing of the 2030 target is notable for its relative restraint: the operative word is “major” (主要), not “primary” or “dominant.” Subsequent policy has built on this foundation. The 2024 Government Work Report first proposed the “AI+” action concept. In August 2025, the State Council issued Opinions on Deeply Implementing the “AI+” Action, setting a new three-step timeline: by 2027, AI integration across six key sectors with intelligent terminal adoption exceeding seventy percent; by 2030, adoption above ninety percent with the intelligent economy as a major growth engine; by 2035, full entry into an intelligent economy and society. The 15th Five-Year Plan recommendations include comprehensive implementation of the AI+ action as a core strategic direction. TC260 released version 2.0 of its AI Safety Governance Framework in September 2025, upgrading the initial version from September 2024 with added provisions for risk classification, lifecycle management, open-source and supply chain security, and enhanced traceability for AI-generated content.

Digital humans have emerged as one of the most visible manifestations of this AI strategy in public-facing applications. Xinhua News Agency launched China’s first AI news anchor, Xin Xiaohao (新小浩), at the 5th World Internet Conference in Wuzhen on November 7, 2018, using Sogou’s digital avatar technology and modeled on journalist Qiu Hao. A female counterpart, Xin Xiaomeng (新小萃), followed in March 2019 for coverage of the Two Sessions. In May 2020, Xin Xiaowei (新小微) became the world’s first 3D AI anchor, featuring multi-camera angles, muscle modeling, and changeable outfits. CCTV deployed Lingyu (聆语), an AI sign language translator, for the 2022 Beijing Winter Olympics. Government service deployments have proliferated. Shenzhen’s Futian District introduced seventy AI digital intelligence employees in February 2025 covering 240 government service scenarios across eleven categories, powered by DeepSeek R1, with public document correction accuracy exceeding ninety-five percent and complaint dispatch accuracy rising from seventy to ninety-five percent. Beijing’s Fengtai District deployed digital human kiosks by Shiyou Tech (世优科技) for round-the-clock guidance in government service halls. Yicheng in Hubei province launched a pilot in July 2024 where digital humans handle 304 high-frequency service items spanning social security and medical insurance. Beijing Internet Court introduced China’s first AI Virtual Judge for twenty-four-hour online guidance, while the Supreme Court in 2022 established an Auxiliary Trial Principle stipulating that AI assists but never replaces judges.

In education, digital human deployments remain at an early stage. East China Normal University developed LLM-driven digital virtual students for teacher training that simulate classroom interactions across subjects, a project selected as a national “AI + Higher Education” typical case by the Ministry of Education. Tianjin Ren’ai College integrated digital human teacher clones with a smart course platform for university physical education instruction in October 2024. Northwestern Polytechnical University deployed digital human factory systems for a course on self-media creation and art practice in 2024. The Huaqiyun Digital Teacher Course Platform serves hundreds of institutions including Zhejiang University and Beijing Institute of Technology, creating one-to-one AI replicas of instructors’ appearance and voice for course video production. iFlytek (科大讯飞) deployed virtual digital human teachers at kindergarten science festivals in Zhuji in June 2025 and in AI labs at primary schools in Shenyang. NetEase Youdao launched “Hi Echo,” a virtual person oral English coach, in July 2023. Government policy has moved in parallel: Beijing introduced mandatory AI classes in approximately 1,500 schools beginning fall 2025, starting from age six, and national targets call for AI education in all primary and secondary schools by 2030 and full integration into textbooks, exams, and classrooms by 2035. Most education deployments to date, however, serve purposes of automated course video production through teacher clones, teacher training simulations, and AI chatbot companions on devices. Full interactive digital human teachers conducting live classes remain aspirational rather than operational at scale.

Healthcare has seen more substantive digital human deployments. JD Health launched what it described as China’s first batch of digital doctors in early 2025, with AI avatars of real medical experts powered by the Jingyi Qianxun (京医千询) medical large language model, which ranked first on MedBench with a score of 96.1. The system handles over thirty-five million user sessions per year, with a reported human replacement rate exceeding ninety percent for routine consultations and ninety-one percent satisfaction. Over eighty percent of JD Internet Hospital consultations use AI services. Aier Eye Hospital introduced Eyecho in October 2024, a high-fidelity digital human using AierGPT, a retrieval-augmented ophthalmic large language model built with over one hundred clinical experts, deployed across more than nine hundred Aier hospitals globally for pre-surgery education, post-surgery companionship, and bilingual virtual hospital guidance. Guangxi Medical University First Affiliated Hospital launched China’s first hospital-owned-IP urology AI doctor on DeepSeek architecture in February 2025, trained on over three million characters of urology corpus and handling more than eighty percent of standard inquiries. Tsinghua University published research in May 2024 on an Agent AI Hospital comprising fourteen AI doctor agents and four nurse agents across twenty-one specialties, achieving 93.06 percent accuracy on the MedQA dataset, with clinical pilot integration begun at Beijing Tsinghua Changgung Hospital in April 2025. Gushengtang (固生堂) developed a Master TCM AI system based on thirty years of clinical data from ten traditional Chinese medicine specialists across eighty clinics in China and Singapore. DeepSeek was deployed in over 260 hospitals across 93.5 percent of provinces as of May 2025. The National Health Commission mandated all tertiary hospitals integrate AI-assisted diagnosis by 2025. Most deployments, however, serve functions of patient education, triage, navigation, and science popularization rather than autonomous clinical diagnosis, and AI systems explicitly operate as assistants to human doctors.

Entertainment has produced China’s most commercially prominent digital humans. Luo Tianyi (洛天依), launched on July 12, 2012 by Shanghai Henian Information Technology as the first Chinese VOCALOID voice bank, has accumulated over seven thousand original songs and approximately five million Weibo followers. Her milestones include China’s first virtual singer appearance on the CCTV Spring Festival Gala on February 11, 2021, and a sold-out solo concert in Shanghai in 2017 that drew over seven thousand fans with tickets selling out in two minutes. She is regarded as China’s most successful and first profitable virtual digital human IP. AYAYI, launched on May 20, 2021 on Xiaohongshu by RanMai Tech (燃麦科技), became China’s first hyper-realistic digital human, drawing approximately three million views and forty thousand followers overnight and collaborating with Guerlain, Louis Vuitton, Porsche, Tiffany, and Bose before her parent company pivoted toward broader digital asset marketing. Liu Yexi (柳夜熙), launched on October 31, 2021 on Douyin by Chuangyi Tech (创壹科技), combined virtual beauty content with a Chinese fantasy supernatural theme and attracted 2.7 million likes and 1.35 million followers within twenty-four hours, peaking at approximately 8.8 million Douyin followers. JD.com created a digital clone of founder Liu Qiangdong for livestream e-commerce that generated over twenty million views and fifty million yuan in sales during its first show, powered by JD Cloud’s Yanxi large model.

The industry has experienced rapid commercial expansion alongside significant volatility. Approximately 993,000 digital human-related companies were registered in China as of 2023, with over 400,000 newly formed that year, though sixty to seventy percent of digital human service providers from 2023 had disappeared by mid-2024. Market size estimates vary enormously depending on definitional scope. iiMedia Research reported a core market of 205.2 billion yuan in 2023 and 339.2 billion yuan in 2024, with projections of 480.6 billion yuan for 2025, using a broad definition that produces figures roughly fifty times larger than IDC’s narrower projection of 10.24 billion yuan for China’s AI digital human market by 2026. Investment activity has been substantial. From January to July 2025, twenty-three investment deals totaling 3.507 billion yuan were recorded, nearly matching full-year 2024 volume in the first half alone. Silicon Intelligence (硅基智能), with nine funding rounds from Tencent, Sequoia China, and 360 Group, achieved unicorn status. Xiaoice (小冰公司) raised one billion yuan for its AI Being framework. Baidu’s digital human revenue grew fifty-five percent quarter over quarter to approximately 500 million yuan in the second quarter of 2025. Major platform companies have launched accessible production tools: Baidu’s XiLing (曦灵) platform offers 3D super-realistic digital humans from 199 yuan, Tencent’s ZhiYing (智影) provides AI creation and digital human video tools, and Alibaba’s Tongyi Lab has open-sourced 3D digital human tools. Beijing issued the first national-level digital human industry policy in August 2022, targeting a municipal digital human industry exceeding fifty billion yuan by 2025 with goals of cultivating one to two leading enterprises with revenue above five billion yuan. Shanghai’s Hongkou District offers up to ten million yuan for office space and up to 500 million yuan in computing resource subsidies for the sector.

China has constructed a layered regulatory architecture around AI-generated content and digital humans. The Algorithm Recommendation Provisions, jointly issued by the Cyberspace Administration of China, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration for Market Regulation, took effect on March 1, 2022, addressing algorithm transparency, user opt-out rights, and anti-price-discrimination. The Deep Synthesis Provisions followed, effective January 10, 2023, requiring content labeling and watermarking, real-name verification, and algorithm filing while prohibiting the use of deep synthesis for fake news. The Generative AI Interim Measures, issued by seven bodies led by the CAC, took effect on August 15, 2023, covering training data compliance, content labeling, and safety assessment, with over 748 services filed under the measures as of 2025. AI-Generated Content Labeling Measures were issued in March 2025 for implementation on September 1, 2025, accompanied by mandatory national standard GB 45438-2025. Face recognition security management measures took effect on June 1, 2025. The first national standard for digital humans, GB/T 46483-2025 for customer service applications, was published in November 2025, led by SenseTime. The broadcasting standard GY/T 411-2024 for digital humans in broadcasting was issued in November 2024. The foundational legal framework underlying all these regulations comprises the Cybersecurity Law of 2017, the Data Security Law of 2021, and the Personal Information Protection Law of 2021. On April 3, 2026, the CAC released for public comment the Draft Measures for Digital Virtual Human Information Services (数字虚拟人信息服务管理办法), comprising five chapters and twenty-seven articles. The draft requires consent for biometric modeling, protects the image rights of deceased persons, prohibits providing minors with virtual companions, and mandates continuous display of a “digital human” label.

China’s AI capabilities extend internationally through multiple channels. Chinese technology companies supply AI surveillance technology to at least sixty-three countries according to the Carnegie AI Global Surveillance Index, with thirty-six of those countries signatories to the Belt and Road Initiative. Research from MIT and VoxDev found that China is the largest exporter of facial recognition AI, reaching approximately eighty-three countries through 238 trade deals, with China’s AI export propensity 47.4 percentage points greater than its exports of other frontier technologies. Forty-four percent of autocracies import Chinese surveillance AI. The Australian Strategic Policy Institute reported in December 2025 that China is the world’s largest exporter of AI-powered surveillance technology. Huawei’s Safe City program alone encompasses seventy-three agreements across fifty-two countries and operates in over seven hundred cities globally. The Digital Silk Road (数字丝绸之路), launched in 2015 as the digital arm of the Belt and Road Initiative, has channeled over twenty-two billion US dollars in investment and loans into digital infrastructure between 2017 and 2023, with Chinese companies exporting smart city systems to 106 countries. Digital human technology exports remain more limited and commercially oriented by comparison. Wondershare’s Virbo (万兴播爆) offers AI digital human tools for cross-border e-commerce in eleven languages, Huawei Cloud’s MetaStudio provides a digital human production platform available internationally, and Alibaba Cloud offers digital human solutions on a global basis. The virtual digital human Tianyu (天妅), developed by Yuanyuan Technology (元圆科技), has been positioned as the first Chinese-style virtual digital human for cultural export, accumulating 5.5 million followers. China’s Global AI Governance Action Plan, a thirteen-point roadmap announced by Premier Li Qiang at the World Artificial Intelligence Conference in July 2025, calls for dialogue among the ITU, ISO, and IEC and proposes establishing an international AI governance organization with China as host. The first national digital human standard, GB/T 46483-2025, was published with the stated intent to push Chinese standards to go global.

[Apr 2026]