Paid pennies to train tools of repression: the humans behind Moscow’s state surveillance

Surveillance companies sanctioned for their role in detaining protesters and activists in Russia have used gig workers around the world to train their facial recognition systems, the Bureau of Investigative Journalism can reveal.

NTechLab and Tevian were both sanctioned under the EU's human rights regime in July 2023 for contributing to the oppression and detention of protestors in Russia. The two companies supply software to Moscow’s public surveillance system, one of the most extensive in the world, which constantly scans faces seeking matches with its “watchlist”.

TBIJ, in partnership with Follow the Money and Paper Trail Media, investigated links between the two companies and a gig work platform called Toloka. Run by three companies in the Netherlands, Switzerland and Russia, Toloka is owned by the Dutch arm of one of Russia’s biggest tech companies, Yandex. Since at least 2019, the platform has hosted tasks that let gig workers train facial recognition algorithms for pennies at a time.

TBIJ has found numerous tasks posted on Toloka for both Tevian and NTechLab, including after the invasion of Ukraine. There is also some evidence of tasks being done for the companies after sanctions were imposed.

Russia’s recent crack down on gatherings commemorating Alexei Navalny, the opposition leader who died in a Russian prison in February, was in part powered by its facial recognition technology. At least 19 people were detained after their images were identified by Moscow’s surveillance systems, including a woman who laid flowers in his memory and another filmed at his funeral, according to Russian human rights organisation OVD-Info.

The Kremlin also plans to use NTechLab software in a secret project for a “Video Stream Processing Centre” that will link up and analyse camera networks across Russia, according to the Kremlin Leaks project.

“The Kremlin are very technologically sophisticated in what they're doing. They’re experts at destroying civil society and building an authoritarian regime,” said Dan Storyev, English managing editor at OVD-Info. “They’re trying to use every tool at their disposal to hold on to power and the arsenal that they have is massive.”

NTechLab and Tevian need gig workers for two major purposes: data collection and data annotation. Data collection tasks on Toloka often involved submitting pictures of their own faces, an attempt to bolster the image libraries used to train facial recognition tech. One such task specified that workers should be of “Afro, Latino [or] Asian” heritage, and offered $0.50 for five photos. Facial recognition software generally underperforms on non-white faces, in part because existing data sets skew white.

Data annotation tasks involved drawing boxes around human bodies, or watching short clips to identify what the person in the video was doing. This kind of data labelling trains automated systems that use machine learning to recognise patterns.

Toloka is part of this booming data labelling industry, sourcing cheap workers, often based in the Global South, to train and improve technological systems. TBIJ reporting found workers in India, Turkey, Pakistan and Bangladesh had done tasks for NTechLab or Tevian.

Santosh*, a Toloka worker based in India, was concerned to learn that the tasks he had done for NTechLab might have contributed to politically-motivated surveillance. “Nobody knows how they are training the data and what they will do in future. So it's a great risk for the entire world I think.”

The workers were told little about the point of the tasks, but could often see which company had requested the work. As well as facial recognition companies, Toloka also contracts workers for self-driving car companies, search engines and online shopping sites.

TBIJ found evidence that suggests Toloka has continued to work with NTechLab and potentially Tevian, after they were sanctioned by the EU in July 2023. Two screen recordings of the same task were uploaded to YouTube in September of last year, one of which briefly shows news headlines from September 4. A worker also uploaded a screenshot of a Tevian task to Facebook in December 2023, although it’s unclear if it was a recent image.

Individuals and companies in the EU are prohibited from receiving payments from or providing services to sanctioned entities and the evidence raises questions about the compliance of its parent company Yandex NV.

“If those companies are sanctioned, then making resources available to them is prohibited,” said one EU diplomat. “The moment they receive payments from a sanctioned company, so the moment this transfer enters an EU banking system, it should be frozen.

“The question is… are they doing trade with those sanctioned entities directly? Then it would be a direct violation of sanctions.”

Like other multinationals in Russia, Yandex has been forced to separate its Russian and international assets. An agreement to sell its Russia assets to a consortium of investors is expected to be signed in April.

Yandex NV is the group’s Dutch arm and controls two subsidiaries, one Dutch and one Swiss, that are listed as the companies behind the Toloka platform. (Neither NTechLab nor Tevian has been sanctioned by Switzerland.)

It is listed on the Nasdaq, but trading was suspended shortly after the invasion of Ukraine. About a quarter of Yandex NV is still owned by institutional investors, including the asset management giant Fidelity, the asset management arm of UBS and Norges Bank, Norway’s sovereign wealth fund. Other Western investors appear to have offloaded most or all of their holdings over the past year.

Yandex’s Russian arm said that it had recently created a separate crowdsourcing platform called Yandex Tasks for Russian users and clients. It began migrating Russian clients from Toloka to Yandex Tasks in December 2023. The company confirmed it was still working with NTechLab as of this month.

Toloka told TBIJ that Tevian and NTechLab had a contract with the Russian subsidiary Toloka RU LLC. Toloka told TBIJ: “The volume of work offered through the platform during the transition period [from Toloka to Yandex Tasks] by the two companies was immaterial, both financially and in terms of the number of tasks.” It also said that neither Yandex NV nor any of Toloka’s Swiss, EU or US entities “have ever provided services to or received any payments from NtechLab or Tevian”. Toloka said it does not cooperate with companies that violate terms service rules or are using them for unethical purposes.

Last year, NTechLab told Reuters its software was no longer used by the Moscow metro, but would not comment on whether the rest of the city’s surveillance apparatus was using its software. NTechLab added that its software “is independently managed by the customer”.