China's AI Governance and the Human Face It Wears

China's approach to artificial intelligence cannot be separated from the political system that produced it. The technology is not treated as a neutral toolkit acquired from the market but as an instrument of governance, woven into an architecture of surveillance, social management, and ideological guidance that the state has been assembling for two decades. Within that architecture the digital human occupies a peculiar and revealing place. It is the same apparatus turned outward — the identical capabilities in facial modeling, speech synthesis, and large-language reasoning that power the country's cameras and predictive systems, repackaged as an avatar that greets a citizen in a government hall, walks a patient through cataract surgery, or sings on the Spring Festival Gala. The digital human is where China's AI strategy stops watching and starts serving, and understanding it requires understanding the machine of which it is the friendly face.

That machine's inward-facing side is formidable and well documented. The national video-surveillance program known as Skynet, first deployed in 2005, grew from roughly twenty million cameras in 2017 to an estimated six hundred million by 2025, with eight Chinese cities ranking among the ten most surveilled in the world and capabilities spanning facial, license-plate, and gait recognition. A complementary program called Sharp Eyes extended this coverage from cities into rural counties and villages through a layered network reaching down to the level of individual settlements. The Social Credit System, often misunderstood abroad as a single national score for individuals, is in practice a fragmented set of systems whose center of gravity has shifted decisively toward corporate compliance; most local individual-scoring pilots had quietly ended by 2025. At the apparatus's most intensive extreme, an integrated policing platform in Xinjiang aggregated camera feeds, checkpoint scans, and phone data to flag ordinary behaviors as suspicious, a system reverse-engineered and documented by outside researchers in 2019.

What binds these efforts into a coordinated program rather than a scattering of projects is a single foundational document. On the eighth of July 2017 the State Council issued the New Generation Artificial Intelligence Development Plan, which laid out a three-step timeline: parity with the world's advanced level by 2020, world-leading status in select technologies by 2025, and by 2030 the establishment of China as "a major world AI innovation center," with the core industry surpassing one trillion yuan. The phrasing of that final target is worth pausing on, because its operative word is "major" rather than "primary" or "dominant" — a note of restraint frequently missed by those who read the plan as a bid for outright supremacy. The strategy has since been extended by the "AI+" initiative, formalized in State Council opinions in the summer of 2025, which sets fresh benchmarks for embedding intelligent systems throughout the economy across the years to 2035. It is against this backdrop that digital humans have moved from novelty to instrument of state.

The turn outward began, fittingly, with the state's own voice. In late 2018 the Xinhua News Agency unveiled China's first AI news anchor, a synthetic presenter modeled on a working journalist and built on avatar technology from a domestic search company. A female counterpart appeared months later, and by 2020 the agency had produced what it billed as the world's first three-dimensional AI anchor, complete with muscle modeling and changeable outfits. These were demonstrations as much as deployments — proof that the machinery of surveillance-grade facial synthesis could be pointed at the public as a presenter rather than a watcher. The symbolic logic was unmistakable: the same technical capability that identifies a face in a crowd could also be made to wear one.

From broadcasting the deployments spread into the routine machinery of the state. In early 2025 the Futian district of Shenzhen introduced some seventy AI "digital employees" spanning more than two hundred government-service scenarios, powered by a domestically developed reasoning model, with document-correction accuracy reported above ninety-five percent and complaint-dispatch accuracy climbing sharply from earlier levels. Districts of Beijing installed digital-human kiosks for round-the-clock guidance in service halls; a county in Hubei piloted avatars to handle hundreds of high-frequency requests across social security and medical insurance; and the Beijing Internet Court introduced an AI virtual judge for online guidance even as the Supreme Court affirmed the principle that such systems assist but never replace human judges. The citizen encountering these avatars meets the state not as an object of monitoring but as a recipient of service — the apparatus in its most accommodating register.

Two domains illustrate both the ambition and its present limits. In healthcare, digital doctors built on medical large-language models have been deployed across hundreds of hospitals; one leading platform reported handling over thirty-five million user sessions a year with high satisfaction, and a national mandate called for AI-assisted diagnosis in all top-tier hospitals by 2025. Yet these systems overwhelmingly perform triage, navigation, and patient education rather than autonomous diagnosis, operating explicitly as assistants to physicians rather than substitutes for them. Education tells a similar story of aspiration outrunning reality. Universities have built digital student avatars for teacher training and cloned real instructors to produce course videos, and the state has set targets for AI coverage in every primary and secondary school by 2030, but the fully interactive digital teacher conducting live lessons at scale remains a promise rather than a fact.

The most commercially visible digital humans emerged not from the state but from entertainment, though the state has been eager to embrace them. Luo Tianyi, launched in 2012 as China's first VOCALOID voice bank, accumulated thousands of original songs and millions of followers, became the country's first profitable virtual idol, and in 2021 became the first virtual singer to appear on the CCTV Spring Festival Gala — a public endorsement of a fan-built creation absorbed into the national stage. Beneath such showcases, however, the industry has churned violently. Close to a million digital-human-related companies were registered by 2023, yet an estimated sixty to seventy percent of that year's service providers had vanished by the following summer, and market-size estimates vary so wildly by definition that credible figures differ from one another by a factor of fifty. The public sees the polished avatar; the sector beneath it is a speculative and often short-lived frontier.

Binding the whole enterprise together is a regulatory architecture assembled in deliberate layers. Provisions on algorithmic recommendation took effect in 2022, rules on deep synthesis followed in 2023 with requirements for labeling and watermarking, and interim measures for generative AI brought text, image, audio, and video under a filing regime through which hundreds of services have now passed. In the spring of 2026 the Cyberspace Administration released for public comment a dedicated draft on digital virtual human information services, running to five chapters and twenty-seven articles: it requires consent for biometric modeling, protects the image rights of the deceased, forbids the provision of virtual companions to minors, and mandates a continuously displayed label marking the avatar as a digital human. The state is learning to regulate a moving target with steadily increasing precision, always anchoring the exercise to the same commitments to security, order, and value alignment that govern the apparatus as a whole.

The ambition does not stop at the border. China has become the world's largest exporter of AI-powered surveillance technology, supplying such systems to dozens of countries — many of them partners in the Belt and Road Initiative — through the infrastructure of the Digital Silk Road. The export of digital humans is by comparison modest and frankly commercial, carried by cloud platforms offering avatar-production tools to cross-border merchants in a range of languages. But the larger play is normative: through a global AI governance action plan advanced in mid-2025 and active participation in the international bodies where technical standards are written, China seeks not merely to build these systems on its own terms but to help write the rules by which others build theirs. Seen whole, the digital human is neither a toy nor a sideshow. It is the approachable face of an integrated governance machine — the point at which a strategy conceived for surveillance and social management turns to meet the citizen, and smiles.