top of page

"Oculus" — the Eyes Of Roskomnadzor Censorship

A new scope of censorship - a system is being introduced in Russia to control memes, cartoons and political videos.


We tell you how, using the "Oculus" system, Roskomnadzor plans to find criticism about Putin even in pictures and videos. The system has already been put into operation at the beginning of 2023. We analyze in detail how it works now and what it can do.



 

Since the war in Ukraine, Roskomnadzor has been rapidly expanding the degree of control in the Russian-language Internet and introducing new systems for strict censorship. For example, since 2023, the agency has launched "Oculus". With the help of this system, more than 200,000 images a day are automatically analysed and the so-called banned content will be found in them promptly. This includes propaganda of suicide, drugs, and tobacco products.


Screenshot from the document "Функциональные_требования_ОКУЛУС_11_02_2022.docx"
Screenshot from the document "Функциональные_требования_ОКУЛУС_11_02_2022.docx"

However, the system is primarily designed for political purposes. The system intends to be used to identify calls for mass protest actions and insults to Vladimir Putin in images and videos in a short period of time.


Screenshot from the document "Презентация_по_результатам_НИР_Окулус_24_02_2022"
Screenshot from the document "Презентация_по_результатам_НИР_Окулус_24_02_2022"

The General Radio Frequency Centre (GRFC) is responsible for the project. Employees of the centre found contractors to create “Oculus” and supervise its implementation, use and subsequent fine-tuning. Thanks to the hacking of the servers of the General Radio Frequency Centre, Cyberpartisan obtained the internal documents and source code of the system in the autumn of 2022. And today we'll tell you how “Oculus” works and what its implementation will mean for the Internet.

 

"Oculus" — From Idea To Implementation


Work on “Oculus” began on October 21, 2021, with a scientific study assigned to staff at the Moscow Institute of Physics and Technology.


At a cost of 15 million Russian rubles, they were tasked with finding out in two months whether it was possible to use artificial intelligence to find certain content in images and videos. In addition, MIPT employees had to develop the concept of "Oculus".


As a result of the conducted work, it has been confirmed that it is possible in principle to use artificial intelligence for such tasks. And in December 2021, the technical specifications were issued, according to which contractors searched for the development of the "Oculus" system.


In August 2022, the task was contracted to Execution RDC for 58 million roubles.


Screenshot from the document "Письмо перенос сроков 1 этап.docx"
Screenshot from the document "Письмо перенос сроков 1 этап.docx"

This company was established in 2014 by two first-time businessmen, Artem Ponomarev (on the left below) and Sergei Brailko (on the right below), with a share capital of just 10,000 rubles.



And at the beginning of 2022, it had only 29 employees. They had not previously dealt with government contracts, but they undertook to develop “Oculus” from scratch by December 12, 2022, that is, in just four months.


Screenshot from the document "Команда Окулус 250822.xlsx"
Screenshot from the document "Команда Окулус 250822.xlsx"

According to experts familiar with the technical specifications, the budget and time frame for the project are unrealistic and perplexing. However, in February 2023 Roskomnadzor announced the launch of the system.


Now let's examine what a small company managed to accomplish in such a short period of time and with a modest budget. Let us tell you more about how “Oculus” works.

 

Sources From the USA


The system is based on Yolo 5, an open-source artificial intelligence architecture developed by Glenn Jocher, the founder of the American company Ultralitics.


Glenn Jocher, photo from ultralytics.com
Glenn Jocher, photo from ultralytics.com

You can easily see how Yolo 5 works on the company's official website. All you must do is upload a photo to the website and, online, the artificial intelligence will identify all the main objects it depicts.


Example of YOLO5 working from the website https://ultralytics.com/yolov5
Example of YOLO5 working from the website https://ultralytics.com/yolov5

In terms of neural networks, this project was ideally suited to the implementation of the “Oculus” prototype. Thus, the Russian company Excursion DC has in fact further developed a ready-made solution created in the USA for recognizing various objects in photos and videos.

 

How Oculus Censorship Works


“Oculus” can identify predefined images, such as logos, coats of arms, symbols, and faces of politicians. Below you can see examples of images where the system's artificial intelligence has been trained to identify negative information.



All such images considered violations by Roskomnadzor's censors and should be blocked so that even on the Internet people cannot share their point of view and joke about sensitive political topics.


The system itself does not search for infringing images and videos, nor does it have a user interface at all.

“Oculus” is integrated into a single information analysis module used by Roskomnadzor censors. Through it, images and videos from sources that are already under Roskomnadzor's constant scrutiny are fed into the system. For example, from the pages of famous bloggers, politicians, media personalities and online publications. And “Oculus”, in turn, must analyse these materials and give the operator the result.

 

Better In Words Than In Deeds


The system checks exactly the images. That is, the “Oculus” splits all the uploaded videos into separate frames and searches each of them for prohibited content. And for the task at hand, the system works rather slowly. If we take the speed of 200,000 images per day announced by Roskomnadzor, it takes about 3 seconds to process one image. And even a relatively short video contains several thousand frames. So, the system is unlikely to be able to process more than a few hundred videos a day, while there are thousands of them on the Internet every day.

In other words, “Oculus” will only help Roskomnadzor check a very small proportion of new videos from sources that are already under the censors' control.

And now, the system only searches for specific images. If a video, for example, agitates to join the Right Sector, but does not have its logo or is drawn in a non-standard way, “Oculus” will not recognize banned content.😉


In this way, the system works faster than a human, but not more effectively. And despite loud statements about the launch of “Oculus” in February 2023, now Roskomnadzor has at its disposal, in our estimation, a prototype of the desired solution rather than a full-fledged product.

To create a software package capable of finding banned content in more videos, according to various estimates from internal documents of the State Radio Broadcasting Centre, between 500 million and 1.5 billion roubles is needed. And at the moment it has not been implemented.


Screenshot from document "Презентация_по_результатам_НИР_Окулус_24_02_2022"
Screenshot from document "Презентация_по_результатам_НИР_Окулус_24_02_2022"
 

Napoleonic Plans


According to the internal documents, the plans for work on “Oculus” are scheduled to run until 2025:

  • 2023 — the system is going to "learn" to process videos with broken frame sequences and pictures, as well as improve text recognition and expand the list of objects to search.

  • 2024 — they will add analysis of emotions, the age of faces, and people's poses and actions.

  • 2025 — “Oculus” will be able to distinguish fakes in images and identify people by their silhouettes.

In a developed democracy, such neural network and artificial intelligence technologies would be used primarily for the public good. For example, to detect and block the propaganda of suicide, drugs, paedophilia. But in Russia at the moment, they are mainly used for total political control of the society. And in the future, unfortunately, similar technologies may be used in Belarus.


Therefore, we ask to spread this video so that as many people as possible could learn about “Oculus” system and to publish internal documents on it which we obtained by hacking into the networks of the General Radio Frequency Centre.


Subscribe and follow updates on CyberPartisan's Telegram channels.

Commenti


bottom of page