From Human Proxies to Autonomous Adaptation: The Next Phase of Russia's Hybrid Activity

In the aftermath of the failed early August sabotage operation at Germany’s Leipzig airport, The Insider (an online publication specialising in investigative journalism) published an investigation alleging some of the actors in a wider network of operatives behind the GRU-linked attack: Oleg Levushkin as organiser, Andrei Karshakov as the hand behind the ‘dud drone’, and Roman Kovalevsky as an associate whose role remains unclear. Leipzig is one of an increasing number of incidents through which Russia-linked operations make two emerging features in so-called ‘hybrid activity’ explicit.

Disposable Agents and Disruption

First, as Julianne Smith, the former US ambassador to NATO, has argued, Russia appears increasingly willing to conduct operations in which its involvement is difficult to deny. The old proxy model preserved plausible deniability, which, now, while useful, is no longer a prerequisite. A new organisational model reflects a wider category of inexperienced and untrained operatives whose significance is structural. Disposability allows for replacement and a preservation of capacity. The model also has a potential strategic benefit. Establishing state responsibility does not necessarily settle what response is warranted. Russia appears willing to tolerate some level of exposure so long as the model itself can shift uncertainty about an operation’s attribution to its place in a wider campaign and probe NATO’s response threshold.

Second, disruption offers Russia another vector through which it can impose costs when its efforts to deter European support for Ukraine through other means – such as repeated threats of targeted nuclear retaliation – no longer appear credible. Imposing costs through disruption is not a direct substitute for deterrence; instead of changing foreign policy, disruption redirects resources and increases the costs of an ally’s infrastructure resilience. This risks compounding the political cost of sustained support. The strategic payoff from this type of disruption and sabotage across domains lies in the accumulated costs imposed on NATO Allies, rather than the success of any particular operation.

Together, this organisational model and the use of disruption preserve Russia’s optionality: Moscow can vary its threshold-probing disruption operations, and impose costs without committing to a single course of action or risking direct confrontation, while exploiting differences among NATO Allies’ perception over the significance of its activity and the response it warrants. This is where a fourth member in the wider operational network holds promise for extending such a strategy: Claude.

An Adaptive Loop

Anthropic’s September 2026 threat-intelligence report describes a Russian-linked espionage actor using Claude in AI-driven workflows to conduct intrusions against Ukrainian and European government, diplomatic and defence targets. Agents monitored whether deployed malware was being detected and, when security products identified it, autonomously modified and rebuilt the tools before redeploying them. This was not a fully autonomous campaign: human operators retained direction over the operation, including important decisions about targeting. Separately, Anthropic identified Russia-based actors using Claude Code to develop and test software for an autonomous drone swarm, including target-selection and engagement functions. Together, the instances demonstrate an adaptive loop in which a defensive response no longer necessarily forces an attacker to stop and devise a new approach manually.

This changes the cost of sustaining probing activities in efforts to disrupt and sabotage. Under a more human-dependent model, each new attempt requires time, technical expertise and operator attention. An agent pursuing a human-supplied objective could instead test alternative approaches, assess the results and adjust its behaviour continuously. Applied across multiple target environments, discovering and maintaining pressure points could become cheaper, continuous and adaptive, even when individual attempts are detected or disrupted.

On 13 October, MI5 chief, Ken McCallum, alluded to these risks separately: on the one hand, Russia is turning to criminal and volunteer intermediaries as its traditional covert network has degraded. On the other hand, terrorists were caught seeking to use AI to develop reconnaissance and attack methods. There is no evidence that Russia is already using autonomous agents to manage a coordinated campaign designed to manipulate NATO’s response thresholds. But the operational conditions for such a strategy are emerging. Autonomy could extend Russia’s existing optionality from the choice of actor and operation to the execution of the operation itself. An agent need not understand the political significance of its actions, or act out grievance. It would be enough for it to adapt technically, generating different effects across systems while leaving the state to decide how to exploit the resulting uncertainty.

Implications for NATO

For Russia, autonomous systems would not merely be a cheaper or faster assistant in its disruption effort. Continuous operational adaptation will change the strategic value of optionality itself. For NATO, this reinforces the need to assess persistent patterns of activity rather than treating every incident as a discrete technical problem. Disruption can remove access and force an adversary to expend resources rebuilding its capabilities, but it may not deter further activity if adaptation becomes cheaper. Responses need to connect technical disruption with intelligence cooperation, law enforcement, financial measures and other instruments that impose costs on the networks enabling these operations. Many of these measures are captured in RUSI’s new deterrence framework. It is up to NATO Allies to solve the underlying technical, strategic and political differences in perception. Even when attributions have been made, an operation may produce different effects across Allied networks, vary in severity and interpretation, and be attributed to the same wider campaign only over time.

Moscow has not adopted the use of autonomous systems as a deliberate strategy yet. Greater autonomy need not necessarily produce more effective coercion. Agents can fail, behave unpredictably or generate activity that has little strategic value. Nor does autonomy remove the need for human decisions about objectives, target selection and acceptable risk. But by lowering the cost of technical adaptation, autonomous systems could make it easier to sustain the kind of varied, persistent pressure from which Russia already derives strategic value. Claude will not take Oleg, Andrei or Roman’s jobs, but it could change what the network around them is able to do and when. NATO Allies must pre-empt this challenge, and prevent the capability of continuous adaptation from becoming a challenge of enduring advantage.

© Ahana Datta Fasel, 2026, published by RUSI with permission of the author.

The views expressed in this Commentary are the author's, and do not represent those of RUSI or any other institution.

For terms of use, see Website Terms and Conditions of Use.

Have an idea for a Commentary you'd like to write for us? Send a short pitch to commentaries@rusi.org and we'll get back to you if it fits into our research interests. View full guidelines for contributors.