When algorithms police the future

Artificial intelligence is beginning to alter one of the most fundamental assumptions underlying modern criminal justice. For centuries, legal systems have operated on the premise that individuals are investigated and judged for actions already committed. Predictive policing, broadly defined as the use of algorithms to analyze large datasets in order to forecast where crimes may occur, who may be involved, and which behaviors warrant closer scrutiny before offenses take place, introduces a different logic.

The technology has spread rapidly. AI systems are now used to analyze surveillance footage, identify criminal networks, detect financial fraud, and forecast crime hotspots. Yet as predictive capabilities become more sophisticated, concerns have grown that the same tools that promise more effective law enforcement could also expand the reach of state surveillance. In their May 2026 report China's Surveillance Moves Overseas, the Vanderbilt Institute of National Security argued that advances in AI are allowing authorities to process and synthesize information at a scale that previously required extensive human analysis, making it possible to identify perceived threats before dissent emerges.

As governments acquire greater capacity to infer future behavior from present data, the implications extend well beyond China. What is ultimately at stake is not simply the accuracy of algorithmic prediction, but the extent to which modern states should be permitted to govern on the basis of probabilities rather than actions.

Beyond surveillance cameras

China's surveillance capabilities rest not on any single technology, but on the integration of vast quantities of information into interconnected systems. According to the University of Oxford's AI and Criminal Justice Project, Chinese authorities have spent years linking financial records, telecommunications data, internet activity, biometric identifiers, and location histories in ways that allow information collected by one part of the state to be combined with data gathered elsewhere.

Among the most visible components of that ecosystem is Skynet, a nationwide video surveillance network associated with urban public security and city management. But the significance of such systems lies not simply in their scale. Their real power stems from the ability to convert countless streams of information into data that can be aggregated, cross-referenced, and analyzed.

Police Cloud, introduced in 2015, was designed to consolidate information from banks, transportation networks, mobile phone providers, and government agencies into unified records. In Xinjiang, where Chinese authorities have subjected the predominantly Muslim Uyghur population to extensive surveillance, the Integrated Joint Operations Platform (IJOP) sought to identify what officials classified as abnormal behavior. It incorporated information ranging from fuel purchases and electricity consumption to travel patterns, smartphone applications, and social relationships in an effort to flag potential risks.

The result is not merely a larger surveillance apparatus, but a different conception of surveillance itself.Information that once existed in separate silos can now be aggregated and analyzed to identify patterns and relationships difficult for human investigators to detect. In that sense, artificial intelligence is enhancing not simply the state's capacity to observe, but its ability to infer.

The promise and limits of prediction

Predictive policing rests on the assumption that algorithms can identify patterns in criminal activity faster and more consistently than human investigators.

Experiments with predictive policing have not been confined to authoritarian states. In the United States, for instance, the Los Angeles Police Department adopted PredPol in 2011, a system that relied on historical crime data to forecast locations where offenses were most likely to occur. Chicago similarly developed the Strategic Subject List, an algorithmic tool designed to assign risk scores to individuals based on arrest histories, known associates, and previous interactions with law enforcement.

Neither program survived. Los Angeles abandoned PredPol in 2020 following criticism over racial bias and concerns regarding its effectiveness. Chicago shut down the Strategic Subject List the same year after an inspector general's review concluded there was little evidence that the program reduced violence and warned that it risked reinforcing existing inequalities.

The difficulties encountered by both programs pointed to a deeper challenge.  Historical policing data are not neutral records of crime. They are records of where police have historically concentrated their attention. Increased surveillance generates more arrests and more recorded offenses, producing datasets that can reinforce the same patterns and justify further intervention.

From crime prevention to political risk

If predictive policing seeks to estimate criminal behavior, disclosures from the past year suggest that China's ambitions may extend beyond crime itself.

In October 2025, an anonymous source leaked internal documents belonging to Geedge Networks, a Chinese cybersecurity company whose products are used to help enforce the country's internet controls. The material included descriptions of artificial intelligence systems intended to identify individuals considered likely to become political dissidents before they engaged in overt opposition activity.

Drawing upon internet activity, location histories, and social networks, the systems sought to construct behavioral profiles designed to support what Chinese authorities refer to as social stability objectives, offering a glimpse into how advances in artificial intelligence could transform surveillance itself. Information that once existed in separate databases and required extensive human analysis can now be synthesized automatically into profiles capable of identifying patterns and generating assessments about potential risks.

The significance of those capabilities lies less in their scale than in their application. Using algorithms to identify areas vulnerable to burglary or fraud differs markedly from using them to estimate whether individuals may one day become political risks, raising the prospect of intervention based not on actions already taken, but on assessments of future intentions.

That evolution reflects a broader shift in the use of artificial intelligence. Analysts at the Australian Strategic Policy Institute, a defense and strategic affairs think tank focused on, have argued that AI is enabling Chinese authorities to move beyond broad-based monitoring toward what they describe as "precision governance," allowing officials to direct attention toward specific individuals and groups with greater speed and accuracy.

A global debate with no clear answers

The implications of that shift are not confined to China, as debates surrounding predictive policing have expanded beyond questions of accuracy and effectiveness to encompass concerns that algorithmic systems may reinforce existing inequalities .Amnesty International UK's 2025 report Automated Racism warned that algorithmic policing tools can amplify inequalities when trained on biased datasets. Communities subjected to greater police scrutiny inevitably generate more criminal records, creating feedback loops that reinforce existing patterns of surveillance and intervention.

Artificial intelligence can process information on a scale beyond human capacity and identify relationships that investigators might overlook. Yet risk assessments are not substitutes for evidence, and statistical correlations cannot fully capture human intentions. As artificial intelligence expands the state's capacity to infer future behavior, the boundary between preventing crime and governing risk may become increasingly difficult to distinguish.