appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Zhang Tong

Chinese police scientists have offered a rare insight into how they were able to break into the dark web using artificial intelligence to decode underground content.
The dark web – a realm unfamiliar to most – is built on encrypted communications and anonymous routing technologies that render user activity nearly untraceable, making it a haven for illegal activity and data trafficking.
In recent years, researchers both in China and abroad have made progress in analysing dark web content, but most of those efforts have concentrated on English-language material and little research has been published about content in Chinese.
But in May, a research team from the Criminal Investigation Police University of China published a paper in the Journal of Chinese Computer Systems about a model specifically trained to translate and assess dark web content in Chinese.
The first challenge was infiltrating dark websites to collect data. The researchers found that login requirements varied widely across platforms, and traditional web crawlers struggled with Captcha verification and the constantly shifting anti-scraping tactics adopted by these sites.
To overcome this, the researchers designed a system to handle different tasks such as logging in, making lists and downloading images.
The system also regulated access frequency to evade detection mechanisms.
Ultimately, it gathered more than 10,000 pieces of data from eight Chinese-language trading sites and forums, the paper said.
The text and images were then fed into another model, which converted all images – such as those of bank card details – into unified text descriptions.
Next, the team manually labelled 1,800 samples as “seed” data that was used to train an initial classification model.
The model then employed active learning to label the remaining data, before it was passed to human reviewers for verification.
After six rounds, the model had classified 12,467 samples under eight categories, including the trade in financial credentials, trafficking identity data, offering illegal services, sexually explicit material and extreme or violent content.
Beyond these category labels, each piece of suspect content was further classified depending on whether it was a “minor violation” or a “serious crime”.

Given that most dark web samples are high-risk, the team devised a scoring mechanism with material that contained no sensitive or illegal content being graded 0-0.2. Potentially risky content – featuring sensitive phrases but nothing illegal – was graded between 0.2 and 0.4, while illegal content scored between 0.4 and 1.
The precise grade given to illegal content depended on its nature, with 0.05 points being added for identity theft or references to providing illegal services. Meanwhile, references to minors or the provision of criminal know-how or tools were given a further 0.1 points each.
The list of sensitive words was initially compiled manually based on words and phrases linked to illegal activity on the dark web. This lexicon was then updated based on the model’s predictions, although the paper did not give details of how this was done.
The finished model was validated by an independent test set of 200 samples, which concluded that the AI tools could reliably categorise content found on the dark web.
The paper did not mention whether the model would be used by law enforcement agencies to monitor posts on the Chinese dark web.
“The method can stably perform risk-annotation tasks at limited human cost, providing reusable data and methodological support for Chinese dark web risk analysis,” the team wrote in the paper.
“Future work will expand the data scale and explore more downstream analytical tasks tailored to the Chinese dark web landscape.”


T. rex was warm-blooded and had body temperature close to ours, study finds

China uses 100,000 home-grown AI chips to build leading weather forecast system

Geologists find unexpectedly large water sources in Xinjiang beneath China’s largest desert

Chinese scientists build world’s smallest CT scanner, weighing just 6kg

Science

How Chinese police infiltrated and monitored the dark web with the help of AI tools

The Chinese-language dark web has largely been overlooked by researchers until now. Photo: Shutterstock

The AI model was able to classify dark web content to determine whether it was illegal. Photo: Shutterstock

AI homework tools cut exam scores by 20%, study of 26,000 Chinese students finds

By comparing AI users with non-users, researchers from Sweden and Hong Kong found a clear gap: AI boosted short-term efficiency but harmed long-term learning. Photo: Shutterstock

Will AI kill English classes? Chinese university to drop them as tech booms

Shenzhen University of Advanced Technology in southern China has announced it will do away with standard English courses in favour of classes that emphasise cross-cultural skills. Photo: SUAT

Real-time translation has become “impressively effective”, according to SUAT provost Zhao Wei. Photo: WeChat

Why data may become China’s most durable advantage in the AI race

China sees language resources as strategic infrastructure that underpins the country’s AI ambitions. Photo: Shutterstock

News

China

Researchers linked to law enforcement collected data and devised tools to help classify sensitive and illegal material


Researchers linked to law enforcement collected data and devised tools to help classify sensitive and illegal material.


Artificial intelligence (AI)

Crime in China

How Chinese police infiltrated and monitored the dark web with the help of AI tools

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}Researchers linked to law enforcement collected data and devised tools to help classify sensitive and illegal material

Researchers linked to law enforcement collected data and devised tools to help classify sensitive and illegal material