China’s Model, Part 3: Identity Resolution, Data Fusion, and Preemption

This is a segment of a six-part series analyzing how the Chinese regime conducts its modern influence operations. Read part I here and part II here.

Commentary

The most important thing to understand about China’s surveillance state is that it isn’t built around a single breakthrough sensor. It’s built around an operating system.

Facial recognition makes for compelling headlines because it feels like a superpower: a camera sees you, the state knows you. But the real advantage is architectural. China has spent two decades building the conduit that turns identification into governance: identity registries, dense sensor coverage, and data-fusion platforms that generate alerts, lists, and actions at scale.

That architecture is why “emotional recognition” and other contested analytic layers matter in China even when the science is disputed. In a system designed to act on weak signals, a noisy inference can still become an operational trigger.

Identity as the Primary Key

Every surveillance regime begins with identity resolution: the ability to tie an observation to a person who can be located, pressured, sanctioned, or coerced. Facial recognition is one path to that end, but China’s governance ecosystem increasingly treats facial data as a credential—an authentication layer linked to state databases.

That linkage is explicit in the Chinese regime’s 2025 facial-recognition security measures. The measures encourage (for identity verification or identifying specified individuals) the use of channels such as the Basic State Population Information Database and a State Online Identity Verification Public Service—a detail that matters because it implies a preferred pathway: face matching routed through state-backed identity infrastructure rather than scattered, siloed vendor databases.

This is the element outside observers often miss. A faceprint alone is a pattern. A faceprint mapped into a population registry becomes a governance tool.

Sensors Are Everywhere—the Question Is What They Feed

China’s camera networks are frequently described through brand names—Skynet, Sharp Eyes. The more important issue is function: coverage and integration. AP reporting has described the progression of projects aimed at expanding surveillance across China, including Sharp Eyes, as part of a broader high-tech effort to stand watch over public space.

Intense coverage does two things. It increases the chance of capturing a face or a behavioral cue. It also enables “before and after” logic: patterns over time, deviations from routine, and association mapping. In the West, similar logics exist in pockets. In China, the ambition is systemic. What we are talking about is profiling.

CSET’s work on data fusion makes the point in the bluntest possible way: China’s surveillance programs increasingly rely on platforms that “make sense” of disparate inputs—programs including Sharp Eyes, the nationwide Police Cloud, and Xinjiang’s Integrated Joint Operations Platform (IJOP).

Sensors are plentiful, and they help justify state action against any behavior that challenges the status quo.

The Fusion Layer: From Data to Action

If you want one case study for how the Chinese model works, Xinjiang remains the clearest window—because the system has been documented with unusual specificity by rights groups and researchers.

Human Rights Watch’s reverse-engineering of a Xinjiang police mobile app connected to IJOP describes a platform that aggregates data about people and flags individuals it deems “potentially threatening,” prompting police attention. Australia’s ASPI Xinjiang Data Project summarizes the same basic dynamic: a policing app feeding into IJOP as part of mass surveillance.

The key issue isn’t whether the system is perfectly accurate. It’s how it’s used. A data-fusion platform turns small, ordinary signals—such as a checkpoint scan, a new phone, a changed travel route, or contact with the “wrong” person—into flags inside a machine built to suspect first and ask questions later. It doesn’t have to be right every time, just produce enough “hits” to keep the system constantly intervening.

This is why the previous piece’s warning about emotion artificial intelligence (AI) matters in a China context. A brittle or contested conclusion becomes dangerous when a platform is built to operationalize those assessments.

Police Cloud: Platform Policing With Chinese Characteristics

Outside Xinjiang, China’s public security modernization is increasingly described in terms of platformization—building modular systems that integrate surveillance components while preserving state control.

A 2026 peer-reviewed paper on Police Cloud argues that the Ministry of Public Security’s shift toward “platformization” triggered a broader change in public security governance, and Police Cloud enabled functional modularity across surveillance elements—an architecture that helps the state retain autonomy and control.

That last clause is crucial. In many countries, platform policing raises fears that private companies are accumulating coercive power, as with social media data collection in the West. In China, the state’s challenge is different: to extract innovation and capacity from industry while keeping the command layer in the state’s hands. Platformization is one answer.

Emotion and Affect: Why Contested Signals Still Matter

Seen in this light, emotion recognition is best understood as an optional analytic layer that can be plugged into an existing action pipeline.

China’s emotion recognition market has been described as a “burgeoning” sector, with public security and education use cases in mind, according to ARTICLE 19’s reporting. Reuters’ summary of that report described applications ranging from interrogation-oriented monitoring to classroom attention tracking.

The ethical problem is often framed as “accuracy.” The operational problem is escalation. In settings where authorities have power over you (interrogations, detention, border screening, “stability maintenance,” school discipline, workplace monitoring), an emotion/affect score doesn’t have to be scientifically reliable to change what happens next. That score becomes a reason to push harder, detain longer, surveil more, or classify a cohort as “high risk.”

This is where collective inference becomes strategically useful. China does not need perfect reads on individuals to benefit from analytics at scale. It can operate on cohort-level patterns: the average response of a demographic slice; a neighborhood’s “temperature”; a campus cluster’s reactions to enforcement; and the emotional intensity of crowd scenes in user-generated video. Group-level emotion is a recognized research direction in the computer vision literature, aimed at estimating collective affect from many partial cues.

Aggregate inference changes what is possible. If you can detect shifts in group response, you can tune your interventions—more pressure here, less there—while keeping the overall machine stable. Additionally, if group behavior is the metric, then individuals start to separate themselves from the group for preservation. This also works in the state’s interest.

The Regulatory Paradox: Privacy Rules’ Alongside Expansive State Capability

A reader might reasonably ask: If China is building such powerful systems, why is it also issuing “protective” rules around facial recognition?

Because the Chinese state often pursues two objectives at once: expand capability while standardizing control. China’s 2025 facial recognition measures were reported by the U.S. Library of Congress as aiming to regulate use and protect personal information, including limiting installation in public spaces to what is necessary for public security and emphasizing governance requirements. CSET’s translation and analysis of the measures help show how they shape the ecosystem while preserving pathways for state-linked identity verification.

This two-objective approach is an effort to formalize the surveillance economy: define compliant use, discipline uncontrolled private collection, and steer identity verification toward state-sanctioned channels.

What the Model Means Beyond China

China hasn’t just built a surveillance architecture; it has assembled a coercive operating system for governance, one that stitches together flawed technologies and turns them into a single, fast‑learning instrument of state power. The design is intentionally elastic. Once the fusion layer is in place, Beijing can bolt on new identifiers—gait, voice, “emotion,” micro‑pattern anomaly detection—without touching the core logic. The result is a machine that never stops training on its population, tightening the loop between what it sees and how it acts, and doing so with a confidence that comes from sheer volume rather than precision.

And Beijing doesn’t keep this machinery at home. The regime’s safe‑city exports double as political technology and data‑extraction infrastructure, feeding sensor streams and operational metrics back into Chinese ecosystems while nudging client states toward China’s definition of “order.” That’s the hinge to TikTok and other recommender platforms, which operate as global distribution engines with real‑time measurement baked in.

China’s domestic model shows what happens when identification, fusion, and behavioral steering become routine. When those systems meet—identity, segmentation, and attention control—the question is no longer whether any classifier is perfect. It becomes how fast a state can learn what moves a population, and how casually it can turn that knowledge into leverage.

Read part I here and part II here.

Views expressed in this article are the opinions of the author and do not necessarily reflect the views of The Epoch Times.