Remain of Shahed 136 at an exhibition showing remains of missiles and drones that Russia used to attack Kyiv on May 12, 2023 in Kyiv, Ukraine. (Photo by Oleksii Samsonov /Global Images Ukraine via Getty Images)
In December 2024 Ukrainian military sources shared with Western media that Russia had started equipping its Shahed-136 strike drones with artificial intelligence. According to Kyiv’s EU Ambassador Vsevolod Chentsov, the upgrade made these drones “smarter” — better able to evade Ukrainian air defenses and identify high-value targets such as energy infrastructure.
This is important because Shahed-type drones, including the Russian version of Shahed (dubbed Geran) and simpler decoy models like the Gerbera, have become Russia’s most frequently used strike drones for daily attacks on Ukraine, and their usage continues to rise. As shown on the figure below, the number of these drones’ deployment has been on a steady growth since they were first used in September 2022.
The “good” news, as much as any news out of Ukraine can be viewed as positive these days, is that a Ukrainian military official clarified to me that the technology currently used in these striking drones is more conventional than it appears, and falls short of creating true autonomous drone swarms.
But even so, the capability is yet another sign that Russia is investing heavily to the idea of AI-powered swarms of drones that can overwhelm Ukraine’s defenses. And if the US and its partners do not take action soon, Moscow may well have this capability in the field within the next few years.
Let’s start with what Russia apparently has put into the Shahed-136. According to the Ukrainian official, it involves two upgrades to existing technologies.
Number of Russian Shahed or Shahed-type drone attacks on Ukraine, by month
Source: Ukrainian Air Force, statistics compiled by the author
First, the enhanced navigation is likely based on computer vision algorithms rather than advanced AI. The fact that these drones can operate without GPS suggests they use the new generation of Digital Scene Matching Area Correlator (DSMAC), a well-known guidance system that compares images from onboard camera to a pre-loaded terrain map in real time. Therefore, this system cannot be considered true autonomous navigation, a critical prerequisite for transitioning to networked drone swarms.
Second, the same computer is likely being loaded with an AI model trained to recognize energy infrastructure. That’s a step toward automated target recognition, but the idea itself is not new to modern striking drones. Still, modern technology allows these models to be trained on smaller datasets and quickly adapt to changing conditions — like new protective anti-drone shelters built around Ukrainian power facilities—enabling onboard software not only to recognize the target but also to pinpoint the most vulnerable spot with high accuracy.
The fact these upgrades do not appear to be true AI as dreamed of (or feared by) military planners, they still do represent a threat upgrade. That’s especially concerning when autonomous navigation merges with automatic target recognition and the drone is authorized to engage the target without human supervision. At that point, it becomes a genuinely autonomous weapon system — potentially deployed in swarms.
These upgrades to the Shahed-136 already being in the field is a major warning sign that Russia is actively testing and developing unmanned swarming capabilities. In fact, public information reveals that there are four key indicators suggesting this is precisely the direction Russia is moving towards.
First, for swarming drones to work, mass production is needed. Russia is already well on its way at the key manufacturing site in the Alabuga Special Economic Zone: According to a CNN report, this factory produced 2,738 Geran drones in 2023, and in 2024 it more than doubled its output to 5,760 units during the first nine months of the year. Serial production has also begun for the so-called “Gerbera” decoys — simpler plywood-and-foam drones that mimic real Shaheds on radar but cost ten times less. Russia planned to build 10,000 Gerberas by the end of 2024, twice the planned Shahed output.
To sustain this surge, Moscow relies on imported components from China. Despite official statements of neutrality, Chinese firms allegedly supply 60 percent of critical drone parts via a network of over 30 companies, fueling the growth of Russia’s drone fleet.
Second, there is evidence that the Shahed are being equipped even now with the necessary components for future AI capabilities. A January 2024 leak of classified documents revealed a new Shahed variant labeled “MS 236,” featuring an electro-optical guidance head — an upgrade displayed in March 2024 during Vladimir Putin’s visit to the Special Technology Center in St. Petersburg. More information came to light in December 2024 following Ukraine’s attack on a Shahed production facility in Alabuga. Defense Intelligence reported the facility was stocked with enough parts for 400 Shaheds, including thermal imaging cameras and navigation systems.
Equipping drones with these sensors allow the drone to do mid-flight reorientation and not to rely solely on preloaded coordinates. While earlier versions of Shahed drone still depended on an operator, integrating machine vision moves the Shahed closer to autonomous navigation and automated target recognition.
A third step is the way Russia is in the early stages of using both quantity and technological upgrades to unleash coordinated mass attacks, which would be a key tactic for AI enabled drone swarms. Early signs of this approach have already appeared with the “Gerbera” decoys, which reportedly employ mesh modems allowing them to share data and form a sort of a swarm. These modems can relay real-time information such as each drone’s position, operational status, and any detection of electronic countermeasures, enabling the swarm to reroute and remain effective.
Meanwhile, Shaheds equipped with electro-optical sensors can carry out target identification missions in real time, paving the way for a subsequent wave of numerous drones to strike. By combining autonomous navigation, AI-enabled target detection system, and data sharing, Russia is moving closer to deploying large-scale, coordinated drone swarms that require minimal human input.
Of course, to make an autonomous swarm work, Russia needs to get its artificial intelligent software up to snuff, which is the fourth place where we can see clear investments. Russia is doubling down on this front, relying on partner countries to evade sanctions and secure much-needed technology. In March 2024, for instance, Russia and Iran signed a memorandum of understanding on AI development to enhance “technological and cultural dialogue” and to conduct research of AI applications. Additionally, Russian institutions, such as Moscow State University, are collaborating with Iranian universities to develop advanced AI algorithms for robotics, signaling that these international partnerships may soon channel AI breakthroughs into autonomous unmanned systems.
A further indication of Russia’s AI push — even under Western sanctions — appeared in December 2024, when President Vladimir Putin announced the launch of an AI alliance within BRICS. He unveiled the new “AI Alliance Network,” initially bringing together 20 companies from Russia, Brazil, India, China, Iran, and the UAE — a portion of a larger pool of 50 international businesses reportedly ready to join. Despite Western chip export bans, Russia’s AI sector has raised its spending on equipment by 40 percent in 2024, contributing to a 10.5 percent increase in technology output.
Finally, on Dec. 30, President Vladimir Putin ordered the Russian Government and Sberbank, the Russian leading institution and investor in AI development, to collaborate with China on artificial intelligence research. Russian Prime Minister Mikhail Mishustin and Sberbank CEO Herman Gref must submit a roadmap for collaboration by April 30, 2025.
These are all the moves of a player serious about hitting its ambitious goals for AI development, systematically building partnerships to secure both the hardware and software essential to its war machine. Compounding the challenge, Russia shares military technology advancements and battlefield lessons with allies such as China and Iran, deepening the risk of speeding up deployment of fully autonomous AI-driven weapon systems.
To counter this growing threat, the United States must (1) expand export controls and sanctions beyond directly involved Russian entities, encompassing the broader network of middlemen, shell companies, and covert operatives in third countries; and (2) prioritize blocking Russia’s access to AI-related expertise and knowledge by limiting academic and research connections with both state-owned and private Russian institutions, and ensure more stringent visa reviews for individuals involved in sensitive fields.
By combining these measures with incentives for western private AI firms and universities worldwide to avoid partnerships with sanctioned Russian organizations, Washington can more effectively constrain Moscow’s ability to make progress in developing autonomous weapon systems in mass, such as Shahed and Gerbera drones.
Kateryna Bondar is a fellow with the Wadhwani AI Center at the Center for Strategic and International Studies (CSIS). She is a former adviser to the Ministry of Defense and Ministry of Finance of Ukraine and served as a project manager at the National Reforms Council of the Presidential Administration of Ukraine.
