Russia’s AI industry fuels the war: drones, surveillance, and censorship

Russia’s AI policy rests on the concept of “sovereign artificial intelligence”. In practice, this means adapting to sanctions and trying to rework existing Western solutions for its own needs.

Russia is not trying to make progress in fundamental research or to create new services; instead, it relies on applied uses in a number of areas. Its priorities are AI in UAVs, the oil and gas sector, battlefield command, and AI integration into battlefield command and control.

No fully autonomous drones are known so far, although many have elements of autonomy based on AI. Most often these are target recognition and terminal guidance systems. Such functions are found, for example, in the Geran Seeker (Geran-4, Geran-5), V2U, Lancet (Izdeliye-53), Skvorets, Soroka, Sarych, and Mikrob UAVs.

The main obstacle to the industry’s development remains the lack of its own component base and of much of the hardware, in particular graphics processors and chips. Russia has no access to EUV lithography equipment, without which the most advanced chips cannot be made.

At the same time, Russia can still obtain sanctioned electronic parts through third countries. In 2024 alone, and through India alone, more than $416 million worth of computer equipment and servers were shipped to Russia. These channels are costly, though: sanctioned microelectronics cost double the pre-war price.

Russia can make some hardware components itself (microchips, image sensors, infrared cameras, and radar sensors for UAVs). They are supplied mainly by Rostec companies from the Roselectronics and Element holdings.

The main source of sanctioned equipment for Russia remains grey imports through front companies. Although smuggling makes such purchases possible, this approach has a number of limitations. In particular, technologies become available late, so simpler items have to be produced domestically. Still, any Russian technological solution, even a home-grown one, remains heavily dependent on Western microelectronics.

Kontur divided the companies studied into six functional clusters:

Beyond direct military uses, AI in Russia also serves related military purposes. Facial recognition systems, for example, are built into city video surveillance. Officially they are meant for law enforcement, but Russia’s security agencies use them to track men of conscription age and for covert mobilisation.

Russian developers are also testing drone detection systems, and Rostec units are working on “drone swarm” technology, although no practical use of the idea has been recorded so far.

Universities work closely with defence companies. They supply and develop terrain analysis, satellite-free navigation, and computer vision solutions for UAVs, and are mostly not under sanctions. Money flows into the industry through state funds and accelerators, including Skolkovo, the NTI Foundation, ERA Technopolis, the Moscow Innovation Cluster, and Rusnano.

Sanctions coverage is insufficient: 38 of the 106 companies (36%) are not under sanctions in any jurisdiction, and another 27 are sanctioned by only one. Only 9 companies are under sanctions of the EU, the US, Ukraine, and the UK. Such consolidated restrictions are mostly imposed on long-known defence industry entities and their partners.

By contrast, private developers that emerged or reoriented to the needs of the Russian army during the full-scale war are mostly not sanctioned. Among developers of artificial intelligence for drones, ground robots, and weapon systems, 15 companies remain outside sanctions. The situation is even worse in the part of the ecosystem that does not make weapons directly. Of 13 computer vision developers, 9 are not sanctioned. Of 11 state funds and accelerators, 4 have no sanctions, and another 3 are sanctioned by Ukraine only.

The countries of the sanctions coalition act inconsistently. Ukraine has imposed restrictions on 58 companies, the US on 44, and the EU on 33. The UK has sanctioned only 10, and 9 of them were already under sanctions in the other jurisdictions.

Kontur recommends specific steps to restrict Russia’s military AI:

• designate the 38 companies that are not under sanctions in any jurisdiction, above all the 21 companies whose developments are used directly in the war against Ukraine;
• impose sanctions on the state funds and accelerators that finance military start-ups;
• impose sanctions on those identified companies that are already under restrictions in other jurisdictions;
• designate the ultimate beneficiaries, directors, and chief designers of these companies;
• tighten control over the re-export of AI components, including through secondary sanctions against intermediaries;
• require chip makers to verify the location of their equipment as a condition of an export licence: in the US if the Chip Security Act is passed, with a similar requirement introduced in the EU, Japan, South Korea, and Taiwan.

We analysed the activities of 106 companies that build solutions for autonomous UGVs and drones, as well as machine vision systems for facial recognition and language models for data analysis. Those linked to Russia’s military industry include research universities, business accelerators, and investment funds.