Most counter-drone work starts after a drone is already in the air. Radars, acoustic sensors and interceptor crews try to detect, track and kill it before it reaches its target. A new tool from Ukraine and NATO aims to start earlier, by forecasting where and when the drones are likely to come.
Ukraine's Ministry of Defence said on 13 October that the Armed Forces of Ukraine (AFU) and NATO Allied Command Transformation (ACT) have developed two AI-supported tools, according to Janes. The tools were presented by the NATO-Ukraine JATEC at a three-day seminar in Vilnius in September. The first, designated Model No. 13, predicts enemy UAV attacks. The second, Model No. 32, supports the planning of artillery operations.
UASFeed could not reach the ministry's own release on mod.gov.ua, which returned an access error. This article describes it as reported by Janes, Ukraine's military news outlet ArmyInform and The Defense Post.
What Model No. 13 Claims to Do
According to the ministry's account as reported by Janes, Model No. 13 predicts four things about incoming enemy UAV activity:
- Type: which classes of UAV are likely to be used
- Timing: the time windows in which attacks are likely
- Location: the areas likely to be targeted
- Intensity: how heavy the activity is likely to be
Janes reports that the model produces its forecasts near the frontline from three kinds of input: historical data, weather and the battlefield situation. ArmyInform says the model predicts drone intensity, likely areas, time windows and UAV types "with a high degree of probability". It also says the model was tested on real-world datasets, refined, and is now ready for integration into situational-awareness systems.
Janes quotes the ministry's wording that the tool is "ready for integration into the AFU situational awareness systems." That is a statement of readiness, not of deployment. None of the three reports says the model is running operationally yet, and none gives accuracy figures, error rates or details of the validation datasets. The Defense Post says the ministry describes the model as validated on real-world datasets. UASFeed found no independent evaluation of that claim.
Model No. 32: The Artillery Planner
The second tool is further from the field. The Defense Post reports that Model No. 32 ranks feasible artillery plans. A prototype was handed to Ukrainian stakeholders for testing, and wider deployment depends on further validation. In short, Model No. 13 is described as ready to integrate, while Model No. 32 is still a prototype under test.
Who Built It
ArmyInform says both models were jointly developed by NATO ACT and the Armed Forces of Ukraine, with JATEC involved, and were presented at the Vilnius workshop to experts from NATO member countries and organisations, alongside Ukrainian military specialists.
The Defense Post also places the announcement alongside Ukraine's Avengers Labs platform, which lets developers train AI models on annotated footage collected during the war.
The Scale of the Problem It Targets
The ministry's own figures show why it wants better warning. Janes reports that the ministry said on 12 October that Ukrainian forces intercepted more than 4,700 Russian UAVs and missiles during large-scale attacks in September 2026. Over the same month, according to the ministry, Ukrainian forces neutralised more than 48,000 tactical-level UAVs. The ministry's September breakdown, which also covered missile types, included:
- 2,722 Shahed-131/136 one-way attack drones
- 1,935 Shahed-238s
- 1,009 ISR (intelligence, surveillance and reconnaissance) UAVs
- 719 Lancet UAVs
All of these numbers are Ukrainian MoD claims, and UASFeed has not independently verified them. Janes also notes that the ministry highlighted Russia's increasing use of jet-powered one-way attack UAVs capable of flying at around 400 km/h. A drone at that speed gives defenders less time to react.
Why It Matters
Air defence is a resource-allocation problem. Any defender has a limited supply of fire teams, interceptors, electronic-warfare systems and trained crews, and has to spread them across a long front and a large rear. If commanders know a night's raid is likely to be heavy, aimed at a particular area and timed for a particular window, they could in principle move those assets beforehand rather than react once the drones are already in the air.
Faster threats make that warning more valuable. Against a jet-powered drone at around 400 km/h, detection-to-engagement time shrinks, and getting forces into position early may matter more than reacting quickly. A forecast of likely UAV type matters too. A wave of Lancets near the line calls for a different response than a deep Shahed raid or a surge in ISR drones. The Defense Post notes that Russian reconnaissance drones are used to help direct artillery fire.
There are clear limits. A forecasting model trained on historical patterns is only as good as the assumption that the enemy's behaviour will stay consistent. An adversary who suspects its patterns are being modelled has every reason to change them. The public material gives no measure of how often Model No. 13 is right, how far ahead it forecasts, or how much its accuracy varies by region or drone type. Until there is independent data, "with a high degree of probability" is the developer's own description, not a demonstrated result.
The work may also interest NATO. ArmyInform says the ministry noted that Ukraine and NATO are deepening cooperation on advanced IT solutions, artificial intelligence and machine learning, including sharing expertise and developing common approaches to the use of AI in defence.
What to Watch
- Integration: whether Ukraine confirms that Model No. 13 is running inside AFU situational-awareness systems, and at what level of command.
- Validation data: any published accuracy or lead-time figures, or an independent assessment.
- Model No. 32 testing: whether the artillery-planning prototype moves past stakeholder testing toward wider use.
- NATO uptake: whether ACT carries the forecasting approach into Alliance counter-UAS planning or exercises.
