{"id":29203,"date":"2023-01-18T14:44:07","date_gmt":"2023-01-18T13:44:07","guid":{"rendered":"https:\/\/bird.bg\/?p=29203"},"modified":"2024-12-28T23:31:01","modified_gmt":"2024-12-28T22:31:01","slug":"meta-telus-censorship-pro-russian-propaganda","status":"publish","type":"post","link":"https:\/\/bird.bg\/en\/meta-telus-censorship-pro-russian-propaganda\/","title":{"rendered":"Meta, Telus, Censorship and Pro-Russian Propaganda"},"content":{"rendered":"

N.A. leaves her phone in the antechamber, nods to armed guards and enters the gateway with a two-stage opening leading to her office. The office has matte glass and is located in Telus Tower in Macedonia Square. N.A. takes his post – moderator of content in Bulgarian from Facebook. Today it will process 800-900 reports on a computer that disables the ability to take a screen shot. <\/span><\/p>\n

N.A. works in the Dolphin<\/strong> project. This is the name of the department that moderates the content of the social network Meta. It is the most secret, and therefore the security measures are draconian. Other social networks served by Telus are not as mysterious, for example, Chameleon for Snapchat content. The Dolphin name is also part of the secret, but is found in a series of lawsuits against Telus by fired employees.<\/span><\/p>\n

\"\"
Two cases of employees from Telus’ Dolphin department open at the SCC. There are those in other courts.<\/figcaption><\/figure>\n

For each reported content, N.A. must rule on whether or not it meets the standards of the Facebook community. If it gets tough, N.A. consults the Political Commissar. This is another colleague from the department of politics, who is close to the team leader \/team leader\/. <\/span><\/p>\n

But N.A. doesn’t ask often because he has only a few seconds for each report, and he wants to get his monthly bonus for quantity. An average time of 40 seconds is required to process a report, whether it’s a video, post or comment. For toilet – maximum 8 minutes, for any need. <\/span><\/p>\n

Telus has been working as a subcontractor for Meta since 2019, according to interviews of responsible factors at Meta. Telus itself is very discreet about this collaboration. <\/span><\/p>\n

Dolphin’s Bulgarian department is small – a dozen employees. <\/span>He is subordinate to the Russian department, which is the largest.<\/b> Other small departments subordinate to Russian are Georgian, Armenian and those for languages of Central Asia. The Turkish department is large and independent. It employs over 50 people. <\/span><\/p>\n

BIRD.BG reporters spoke to a large number of Telus employees, both former and present, to figure out exactly how moderation works, which has caused a major public scandal in Bulgaria in recent months. <\/span><\/p>\n

The reason for this interest in the activities of a private company is in the public resonance of the results of moderation, which are very controversial. There is a visible silencing of journalists, media and public figures who are critical of Russian aggression in Ukraine and at the same time a reinforcement of pro-Russian propaganda in a social network that in Bulgaria has the status of a mainstream media. A phenomenon that cannot be explained solely by the standards of the Facebook community. <\/span><\/p>\n

Examples of this are numerous and well documented. As early as March last year, weeks after the start of Russian aggression in Ukraine, a petition was organized to the European Parliament against biased moderation of content in the Bulgarian segment of Facebook. <\/span><\/p>\n

The text<\/a> , signed by more than 2,500 people, emphasizes that leading public figures lose access to their accounts and followers because of statements directed against Russia’s aggressive policy, and numerous signals against pro-Russian propaganda messages, fake news, racial and\/or ethnic insults, and even calls for the extermination of Ukrainians remain ineffective.<\/span><\/p>\n

The petition, also sent to Meta and Telus, expresses doubt that “Bulgarian-language moderators often fail to break away from personal political and geopolitical tastes in their workplace.”<\/span><\/p>\n

Almost a year later, the situation not only did not change for the better, but worsened as the number of blocked journalists and public figures increased critically.<\/span><\/p>\n

Informers, Judges and Political Commissioners<\/span><\/h2>\n

Facebook allegedly uses a sophisticated AI-based algorithm to detect and moderate unwanted content. This is probably the case for large and significant languages \/markets\/, but it is very debatable whether investments have been made for small languages such as Bulgarian.<\/span><\/p>\n

In an interview<\/a> with Capital daily, Jakob Turowski, head of public policy for Central and Eastern Europe, said content management “is always a combination of our employees, AI.” and machine-trained tools.’. <\/span><\/p>\n

The AI system works generally as follows. After mass reporting by trolls, Meta’s algorithm automatically blocks the post. A ticket is created that goes to a second, heuristic AI. Heuristic means that the system makes a summary of partial data without detailed analysis. It decides whether to block the post according to whether such posts have been blocked in recent days by a moderator. <\/span><\/p>\n

The appeal of an AI lockdown leads to a review by a human moderator at Telus. Telus, as part of their trade secret, is also said to have some sort of algorithm for filtering tickets against “unwanted persons”. These are local rules for controversial content that are formed by the company again through a heuristic algorithm learning on metadata. The goal is not humane, but commercial: to remove content that makes users use less of the social network.<\/span><\/p>\n

This is where the first serious problem arises because Telus’ local content rules include the preferences of fake thousands of troll accounts that block “bad things” from “bad people.” <\/span><\/p>\n

These profiles like news of missile strikes on Ukrainian kindergartens and apartment blocks, but do not like news about Russia as a terrorist state. A system that is trained on such data cannot give opposite results. <\/span><\/p>\n

Sources of the BRRD, however, say that because the market in Bulgaria is insignificant, automation is not profitable for it. According to former employees of the company, there is no serious filtration system based on artificial intelligence. Everything rests on reporting by users as in the good old days of State Security. <\/span><\/p>\n

The reported content goes to a Telus judge. The judge is usually a young person fluent in English and another mother tongue. He has been training for four and a half weeks to absorb the knowledge of what is permissible and what is not, according to the standards of the Facebook community. A judge must have an excellent memory because he has no right to take notes. There is a “Bible” with the standards, which is over 1,000 pages, but at 40 seconds on average for processing a post, he does not have time to search in it.<\/span><\/p>\n

The judge sees reported content and decides in seconds whether to approve the report or reject it. Sometimes he needs help from a friend, and then he consults the political commissar of the group. <\/span><\/p>\n

For the author of the reported content, the decision of the judge or political commissioner is the equivalent of a ruling: guilty or innocent. Will he be in the lock in the coming days or free? If he is found guilty by the judge, Facebook’s algorithms will punish him. For repeat offenders, the penalties are more severe and can lead to account blocking.<\/span><\/p>\n

The judge sees who the informant is, he sees who is reported. The reporter does not see the informant. The informant can see the result of the denunciation – whether the reported content is downloaded or not.<\/span><\/p>\n

Purely quantitatively, with such a staging of forces, the winner will be one who manages to mobilize a large number of denunciations. That’s what troll farms do – they report en masse. But the forces that have an interest in distorted moderation are not limited to outside interference.<\/span><\/p>\n

Iranian precedent – 5 to 10,000 euros for blocking<\/span><\/h2>\n

There has already been infiltration and bribery of moderators by Telus in favor of a dictatorial regime. A BBC investigation<\/a> alleges that moderators based in Essen, Germany, received money to block anti-ayatollah content in Iran.<\/span><\/p>\n

The BBC quoted a Persian-language moderator of Telus International as saying Iranian intelligence officials offered 5 to 10,000 euros to content moderators to remove posts from prominent journalists and social activists opposed to the Tehran regime.<\/span><\/p>\n

Pastor Saeed Abedini was removed because it was “contrary to community standards on the sexual exploitation of children.”<\/span><\/p>\n

The publication in question says the following:<\/span><\/p>\n

“Arm the people of Iran before it’s too late and the mullah regime kills them all.”<\/span><\/p>\n

Telus denies bribery allegations against its employees. It went so far that Iranians protested<\/a> outside Telus’ offices in Essen against these practices, and three U.S. lawmakers sent a letter<\/a> to Meta and Telus asking for explanations on how it came to be the de facto assistance to the U.S.-sanctioned Iranian regime.<\/span><\/p>\n

Russia, like Iran, is a hostile power that relies on hybrid warfare involving the modeling of public opinion in its favor. It is known that businessmen close to the regime in the Kremlin like Yevgeny Prigozhin finance the maintenance of thousands of fake troll profiles on social networks. <\/span><\/p>\n

Their mass reports have a guaranteed effect in communities using languages for which an artificial intelligence based on heuristics has been developed. <\/span><\/p>\n

According to our sources, however, Meta has not invested heavily in artificial intelligence and machine-trained heuristic tools for Bulgarian and other small languages. For them, the detailed human intervention of moderators is crucial. <\/span><\/p>\n

Key are education, heredity, socio-political attitudes, geopolitical biases and motivation of judges and political commissars. And also to the people on whom their remuneration and careers depend. <\/span><\/p>\n

That’s why it makes sense for a foreign intelligence agency to target the dozen content moderators in one small language who, despite their modest numbers, still manage to process up to ten thousand reports a day. <\/span><\/p>\n

It is even more effective to recruit people from the leadership and the right political commissars – the people who interpret the standards of the community and mentor young judges what is politically correct and what is not.<\/span><\/p>\n

Three “merindjs” associated with Russia. And the son of Agent Sandor.<\/span><\/h2>\n

The BRRD learned about three employees in positions of responsibility in Telus, for whom it has information about special relations with Russia. These are Ivelina Yankova, Alexander Stoychev, and Julian Parvulov. <\/span><\/p>\n

\"\"
Ivelina Yankova<\/figcaption><\/figure>\n
\"\"
Julian Parvulov<\/figcaption><\/figure>\n

Ivelina Yankova<\/b> is a Senior Client Manager Telus International Europe Bulgaria. She is from Varna and has completed her education at the Academy of Labor and Social Relations in Moscow \/Academy of Labor and Social Relations\/. <\/span><\/p>\n

Julian Parvulov<\/b> is a product manager at Telus. His mother, Olga Leonidovna, was Russian.<\/span><\/p>\n

\"\"
Zorica Gajic<\/figcaption><\/figure>\n

Close to them are the senior cadre <\/span>Boris Koedjikov<\/b>, son of the State Security agent “Shandor”<\/a> from the PGU <\/span>Ivan Koedjikov<\/b>, as well as <\/span>Zorica Gajic<\/b>, the daughter of the Serbian pharaoh Miroljub Gajic, who is reported<\/a> to be close to the BSP.<\/span><\/p>\n

Alexander Stoychev<\/b> from Burgas is in a lower position in Telus, but is a prot\u00e9g\u00e9 of the above “merindjii”. He had a privilege unavailable to any other employee. Every month, he took a two-week leave to visit… Russia. A review of his Facebook profile shows that in 2013 he designated himself in the separatist Donetsk.<\/span><\/p>\n

We have not been able to identify unequivocally the head of the Russian department, who, according to insiders, is called Maya and also has Russian roots. Its role, however, is important, as is the current political commissioner of the Bulgarian department. In this regard, we have sent specific questions to the press service of Colpoint New Europe<\/a> Ltd., as the Bulgarian legal entity owned by Telus International (U.K.) Ltd. is called.<\/span><\/p>\n

– What is the hierarchy of teams in the Dolphin project for content moderation from Meta? Is it true that the Bulgarian team is subordinate to the Russian one? Who manages the Russian team? <\/span><\/p>\n

– Do the leaders and members of the Bulgarian and Russian team have ties with Russia – family, educational, emotional?<\/span><\/p>\n

– Do you have data on relations with Russia – family, educational, emotional, management of the company related to the Dolphin project?<\/span><\/p>\n

– Are the following persons related to the Dolphin project, to the determination of the remuneration and bonuses of employees in it or to their career growth: Boris Koedjikov, Ivalina Yankova, Yulian Parvulov, Alexander Stoychev?<\/span><\/p>\n

Telus’ answer reads as follows: <\/span><\/p>\n

“These people have nothing to do with the subject. TELUS International employs people of different nationalities, carriers of different social characteristics. We categorically decline, being a responsible employer, to comment on or provide information about any of our employees.”<\/span><\/em><\/p>\n

We also asked to find out how many reports on content violating Meta standards have been processed by Meta employees since the start of the contract with Meta for a dozen specific individuals – journalists and active public figures, as well as what opinion they have given about the reported content. We asked for a breakdown of the number of reports and the relevant opinions for each person, providing a link to the respective profile for Atanas Chobanov, Dimitar Stoyanov, Hristo Komarnitski, Tatyana Kristi, Georgi Georgiev \/BOETS\/, Daniela Gorcheva, Evgeniy Dainov, Nikolay Slatinski, Ivo Balev, Ivan Chervenkov and Radoslav Bimbalov.<\/span><\/p>\n

We also asked for information on who or which employees judged that the following publications did not meet Meta standards and what exactly they did not meet:<\/span><\/p>\n