Smart policing with Chinese characters:智慧警务
Revisit with us China's big data policing methods which are currently in the process of getting super charged by AI according to a new study.
In light of the recent and excellent report from the Australian research institute ASPI, on how AI is supercharging China’s surveillance state, we at NetAskari thought it would be a good moment to offer a brief primer on how Big Data - or “Smart Policing” - has been implemented in China over the past few years.
Our analysis is based primarily on the document《智慧警务—大数据环境下新时代公安信息化建设模式探索 (Smart Policing: Exploring a New Model for Public Security Information Construction in a Big Data Environment). This is not some secret internal communiqué, but a publicly available publication from China’s top university Tsinghua.
This official 2018 “handbook” outlines how new Big Data collection systems can support national law enforcement in solving crimes, maintaining public order and stability, and generally streamlining their work. To this, we will add some of our own first-hand experiences and recent observations.
As we will see, the definition of a “criminal” in this context is extremely broad, and the categories of information the Ministry of Public Security considers acceptable to collect from citizens are equally expansive. For seasoned China-watchers, none of this will be particularly surprising or entirely new, but we believe it is valuable to pull these elements together in one place. If you are a frequent traveler to the PRC and you are still a little bit in the dark, this might be a good primer for you.
Big country - Big data
Chinese law enforcement and security institutions traditionally always struggled with the fact that they had far more data collected on specific targets, or the society at large, than they could process. Collecting information is one thing, processing it efficiently is another.
Thanks to advances in modern computing, the rapid expansion of data centers, and a broader push to modernize China’s internal and external security apparatus, 2017 marked a major effort to bring Big Data analysis directly into the hands of Chinese police. The goal was to provide fresh, readily accessible information; data that could be queried, monitored through live surveillance, or used to construct predictive patterns.
And with that, “智慧警务” (Smart Policing) was born. This new buzzword helped motivate the notoriously fragmented provincial public security bureaus to digitize and consolidate their datasets, establish continuous data inflows, eliminate redundancies, and eventually share information nationwide. In theory, this would allow every police prefecture to seamlessly access and work with the same data.
Of course, such a transformation does not happen overnight. In practice, this process has been gradual and remains ongoing even today.
Smart platforms
According to the handbook, a key stage in this development was the creation of shared protocols, interfaces, and database schematics; the foundational blueprint for what would later be christened “雪亮工程” (Sharp Eyes Project). The goal was to bring together all forms of surveillance, whether video feeds, data on movement, online messages, social media profiles, or phone call monitoring.
The handbook repeatedly acknowledges that data generated by 1.3 billion people is enormous in scale. Even if all of it can be stored and processed, running searches alone can take considerable time. As a result, much of the manual focuses on how data should be stored, in what format it should be exchanged, and how information-processing pipelines can be efficiently constructed.
An especially interesting detail appears on page 119: every entity in the system, eventually including all 1.3 billion Chinese citizens, can be reduced to a single hash value derived from multiple personal identifiers that don't need to be topically related. So even subjects that lack a national ID for example can be given a clear, one single value identifier. This allows for faster searches and quicker identity checks by the system.
The handbook contains many more elements on how data is cleaned, how basic machine learning is applied to help structure it, and how the resulting information is prepared for downstream use. But we won’t dive into those specifics here. If you’re interested, the full 277-page document is worth a look.
How does it work ?
All of this sounds impressive on paper, but how does the system actually work in practice? It’s important to remember that this exposé is based on a book published in 2018, so we should assume that today’s capabilities far exceed what we are examining here.
How would a regional Public Security Bureau (PSB) interact with such a system? What tools could officers use to locate individuals, track them, or query information about them?
Fortunately, the publication includes a series of screenshots from an example application used by the province of Hunan’s PSB. And as the saying goes, a picture speaks a thousand words…
By the looks of it all this data is available to a single officer with access to this system. And the data is provided 24/7. No specific court order is needed to investigate an individual, just a login and internal system permissions.
From personal experience, we can say that this system often works remarkably well. During reporting trips into “problematic” regions such as Xinjiang or other sensitive areas, police were at times surprisingly quick in pinpointing and tracking us. They were able to track down contacts or people we talked to and sometimes undercover police “greeted us” on arrival at the local train station or airport. The zero-covid policy most likely boosted the capabilities of this system as well.
Garbage in…
Of course, any analytical system is only as good as the data fed into it. In the past, this was one of the system’s major Achilles’ heels: many regional PSB officers lacked the training to input data properly, and automated collection systems were not yet fully established or interconnected.
But as Chinese society has increasingly embraced online and digital services, relying on countless apps and mobile-based platforms, the surveillance system has grown both denser and more efficient. Data may still be delayed, AI-driven facial recognition cameras may fail, but the more sensors and data streams that are hovered up, the tighter the mesh of the surveillance web becomes.
Final thoughts
It would be unfair to place the spotlight solely on China. Big data analysis, data mining, and predictive policing are practices many police forces around the world engage in. In some cases, these tools can genuinely help identify and apprehend real criminals or dangerous individuals and groups.
But when we see “ethnic minorities” listed alongside categories such as “ex-convicts, drug addicts, fugitives”, and “pickpockets”, it becomes clear that this toolset is being used for far more than "just” criminal investigation and prevention. Nowhere else - not counting North Korea - is the interconnection between social control, political objectives, institutions and a one-party strategy as tightly woven as it is in China. When joining the Chinese police (公安) students have to learn the first element of the “sixteen-character guideline for public security work”: "Loyalty to the party”.
Private enterprises are required to feed their data into these systems, with very little room to refuse. As a result, security services have access to extraordinarily fresh and comprehensive data, likely far more than what most Western counterparts can rely on (though the gap is narrowing). This tools are used by the police to keep dissenters in check, ensure the censorship of information and to chase down threats to the parties official narratives.
During our years working as journalists in China, we saw this system evolve into a “hammer for which every problem looks like a nail”. And its scope continues to widen: it is no longer limited to domestic data sources. Over the past months, we have documented multiple cases suggesting that these mass-analysis systems are now ingesting information globally. The system is still not perfect and many loopholes persist and maybe never get ironed out. But it is an impressive demonstration of what a coherent, holistic and interconnected system can deliver in real time.
None of this is of course to excuse the overreach of Western security services, especially when they enter partnerships with companies like Palantir or Dataminr, often under surprisingly limited oversight regarding what data can be processed, stored, or acted upon. Let the situation in China be a warning to societies who fail to pay attention to those developments.
Strangely enough, even today many Chinese are not always aware of the reach and capabilities of the authorities despite them being rather open ( see the publication we base our analysis on). And as we see China expanding its sphere of influence, this might also stretch soon into other societies (see the “Geedge” leaks etc.).
For those needing the pdf version of the qinghua report: https://ia800600.us.archive.org/5/items/ittushu-2470/%E6%B8%85%E5%8D%8E%E5%A4%A7%E5%AD%A6%E5%9B%BE%E4%B9%A6%E9%A6%86-%E6%88%98%E7%96%AB%E7%89%88/D_%E6%94%BF%E6%B2%BB%E3%80%81%E6%B3%95%E5%BE%8B/3206311_%E6%99%BA%E6%85%A7%E8%AD%A6%E5%8A%A1%E2%80%94%E2%80%94%E5%A4%A7%E6%95%B0%E6%8D%AE%E7%8E%AF%E5%A2%83%E4%B8%8B%E6%96%B0%E6%97%B6%E4%BB%A3%E5%85%AC%E5%AE%89%E4%BF%A1%E6%81%AF%E5%8C%96%E5%BB%BA%E8%AE%BE%E6%A8%A1%E5%BC%8F%E6%8E%A2%E7%B4%A2_text.pdf