Half a century ago, in October 1976, when the “Gang of Four” was arrested, Deng Xiaoping’s family and his son-in-law, He Ping, squeezed into the bathroom, closed the door, turned on the tap to its maximum, and whispered to each other.
This intimate scene from the bathroom came from the handwritten memories of Deng Rong, daughter of Deng Xiaoping: “At that time, no bugs were found in the room, but the family couldn’t rule out the possibility of being eavesdropped on, so they only dared to talk by the sound of running water.”
This is one of the issues explored by the exclusive report of The Epoch Times, titled “The Chandelier Has Eyes” – Is the technical surveillance system of the public security really “all-knowing”? Does the technological control of the party-state also work by instilling fear and uncertainty in the public?
The report did not just focus on equipment catalogs or urban legends, but reconstructed a chain of authoritarian power composed of missions, markings, fronts, platforms, sidelines, and materials.
The report categorizes the materials into four types: 1) Physical objects, judicial records, and independently obtained contents directly supporting them are called “facts.” 2) The judgments made by public researchers or judicial authorities are called “source attributions,” and the judgment subject must be retained in the text. 3) The minimum working process abstracted from various facts is called an “analysis model.” 4) Supervision and advocacy claims made by the report itself are separately labeled as “normative claims.” These four cannot be substituted for each other.
This classification strictly distinguishes between “unconfirmed rumors” and “facts supported by objective files or physical evidence.” A dropped magnetic tracker can prove the existence of a tracking function, but cannot directly infer the command chain behind it; a foreign court ruling can prove the operation of international agents, but cannot prove the corresponding authorization basis for domestic public security cases.
The report divides technical facts and agency attributions into two independent chains of evidence, eliminating the need for speculation to fill evidence gaps in research.
How does a technical surveillance system evolve from data collection to actual control of a person? The report summarizes this process as five transformations:
The first step is turning a person into a recognizable target. A living person, once entered into the system, may be broken down into numbers, devices, vehicles, accounts, faces, voices, or addresses, enabling the system to identify, search, and track the same person repeatedly.
The second step is to let surveillance tasks cross different systems. After the public security proposes a surveillance task, it may be executed through data queries, communications monitoring, operating front-end devices located at specific places, or requesting cooperation from other units, among other ways. Communications, roads, platforms, and working units were originally separate systems, but data may be interconnected through these interfaces.
The third step is to trace more relationships from one person. A phone call may reveal the other party; vehicle tracking may show companions; extracting terminal device data may also find past contacts and groups. In other words, the data obtained through technical surveillance often extends beyond the initially targeted individual to people connected with them.
The fourth step is to turn data into on-the-ground actions. Clues found by the system do not automatically translate into surveillance or detention actions. It requires human judgment, local verification, and other steps before it can advance to surveillance, tailing, detention, or strengthened monitoring before specific activities.
The first four steps illustrate how data gradually transforms into action capability, while the fifth step involves another issue: why monitoring processes become difficult to trace.
In the fifth step, original communications, audio-visual records, and location logs may undergo summarization, transcription, or derivative evidence extraction before entering subsequent procedures. When original data is not preserved, or relevant tasks do not enter reviewable procedures externally, it may be difficult for outsiders to reconstruct how the data was obtained, used, and who made the decisions. Regarding politically sensitive tasks, the report points out that relevant activities may not even enter the litigation process.
The report acknowledges that the evidence strength supporting this model varies.
For example, the phone inspections at the Xinjiang border in 2019 is a case in the report that most directly demonstrates how public security technical surveillance operates: passengers were asked to unlock their devices, staff connected the phones to workstations, and ran data extraction programs. In this scene, personnel, equipment, location, and national coercive powers all appear, forming a more complete operational chain.
The SenseNets platform with approximately 6.7 million location records and over 2.5 million identity files within a day also presents another perspective: the data relationships between individuals, places, and times are already operational within the system. However, the report also cautions that the sources of sensors on the platform vary, hence not all events can be generalized as simply “facial recognition.”
As for the vehicle device photos released by the Cheng Yuan family at the end of 2025, features such as magnets, communication card slots, and anti-tamper components are sufficient to support the conclusion that it is a cellular network return locator. However, the report acknowledges a lack of a complete evidence chain – while the technical features of the device can provide strong clues, they alone cannot confirm the agency behind the installation or the data flow.
Among the cases analyzed in the report, the technical details of the Wang Lijun case are the least, but most illustrative of how the approval process failed.
During Wang Lijun’s tenure as head of Chongqing Public Security, he illegally adopted technical surveillance measures on multiple individuals without approval or falsified approval procedures. This part has been confirmed by public prosecution, trial, and judgment. However, with insufficient monitored object information publicly available, it is impossible to determine whether the monitoring was for political purposes or ascertain the specific technologies employed.
What it can prove is another matter: those in command may also manipulate approval processes.
The core conclusion of the report is: “Signing does not equate to independent review.” When the justification for monitoring, approval, and execution of supervision all occur within the same administrative system without the monitored party’s involvement, and external reviews may be delayed, the approval process may become purely formal. The Wang Lijun case shows that those in command could manipulate or circumvent internal approval procedures. However, the report also warns against generalizing based on this case alone.
The value of the Wang Lijun case lies not in proving all political monitoring but in leaving an institutional problem in the public trial materials: when those in power can manipulate the approval process, who ensures the approval process functions as intended?
The report reiterates that portraying the technical surveillance system as all-powerful serves to promote it for intimidation purposes. A microphone can only listen to one space, a tracker can only trace one carrier, and field personnel may also be exposed or lose track – the front-end capabilities are always limited.
However, for those being monitored, as long as they cannot determine which conversations or routes have been included in the system, fear may come into play. Monitoring does not need to cover everyone; as long as people cannot prove they are not being monitored, they may choose to comply. This mechanism is similar to the atmosphere in North Korea where “everyone could be an informant”: actual monitoring resources are limited yet make the cost of challenging the system appear higher than compliance.
Therefore, the report deliberately separates “technically feasible” from “politically used.” To determine if monitoring is for political purposes, at least one supporting evidence from two categories among target selection, task purposes, relations based on cases, and control exports must be obtained, with two independent pieces of evidence required from the two categories. Otherwise, the conclusion can only stay at “political usage-oriented” or “political context.”
This restraint is not to absolve public security but to avoid conclusions exceeding the evidence. If the level of disclosure surpasses the evidence’s support, it may ultimately damage the credibility of the discloser.
The report also admits to a crucial gap: so far, no domestic case has been able to publicly connect the original mission text, full front-ends, complete running logs, political purposes, and results writing back these five steps. This gap should be retained rather than filled with speculation.
Therefore, the report proposes that the key moving forward is to demand public security to leave reviewable records for each technical task, including: factual basis to determine targets, start and stop times of front-end devices, reasons for expanding monitoring scope, human review of algorithm filtering results, and closing records of each interface when the task ends.
Political commentator Li Zhengming stated that,
For those who have long been concerned about the risks in the CCP system, this report truly reveals a question: Is anyone responsible for errors in the public security surveillance system?
If the answer is consistently no, fear may continue to operate – no need for a lamp to actually have “eyes.”
The Epoch Times’ groundbreaking study “The Chandelier Has Eyes – The Insider of CCP Public Security Technical Surveillance” e-book is now available, revealing the operating mechanism of CCP public security technical surveillance and its political utility.