Ukraine’s Ministry of Defense is letting selected defense companies train battlefield AI models on millions of annotated combat-drone frames through Avengers Labs, a Brave1-linked partnership platform that keeps the raw data inside a protected Dataroom and returns the finished models to Kyiv.

The program gives Ukrainian and foreign defense developers access to visual and thermal imagery from real drone missions, according to source material citing Ukraine’s Ministry of Defense, Minister Mykhailo Fedorov and reporting from Reuters, Kyiv Post, Kyiv Independent, Ukrinform and UNITED24. The dataset is described as covering aerial and ground targets in difficult conditions, including night flights, fog, rain, camouflage and electronic-warfare environments.

Companies do not receive raw combat footage to take away. The training work takes place inside the Brave1 Dataroom, described as a secure environment built by the Ministry of Defense, the Ministry of Digital Transformation, the Armed Forces, a military-intelligence research institute and Palantir. More than 100 Ukrainian companies are said to have access, with international partners brought in through Avengers Labs.

The exchange is unusual for defense AI: partners get access to scarce operational data, while Ukraine keeps the improved models they produce. The underlying Avengers system is already tied to Ukraine’s DELTA and VEZHA battlefield-management tools and, according to ministry reporting cited in the source material, flags roughly 12,000 enemy units per week. That figure has not been independently verified in the provided material.

Combat Data Becomes Leverage

The program matters because battlefield computer vision depends less on generic AI capability than on real examples of what the model must detect. Clean test imagery does not match drones flying through jamming, smoke, weather, darkness and battlefield clutter. Ukraine’s front line has produced a dataset that few other states or companies can reproduce.

By controlling access rather than exporting the raw archive, Ukraine is trying to convert wartime data into military capability while limiting data leakage. For defense firms, the appeal is access to rare training material. For Ukraine, the payoff is a stream of improved detection, tracking, navigation and counter-drone models that can be folded back into its own systems.

Training Data for Machine Learning: Human Supervision from Annotation to Data Science

View Latest Price

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

From Drones To Avengers

Ukraine has spent years building a defense-tech system around drones, rapid procurement and battlefield feedback. Brave1, the government-backed defense-innovation cluster, has served as a route for startups and military users to test and field wartime technology faster than traditional procurement systems usually allow.

Fedorov, formerly Ukraine’s digital-transformation minister and associated with the Army of Drones effort, became defense minister in January 2026. The Avengers Labs model fits that broader push: turn wartime digital infrastructure into usable military products, then return those products to Ukrainian forces.

The source material says the current focus includes automatic detection of enemy drones, onboard machine vision for front-line drones, autonomous navigation in GPS-denied skies and Shahed interception workflows in which a human remains responsible for the firing decision.

Limits Of The Data Deal

Several details remain unclear from the available material. The exact terms for foreign companies, pricing, export controls, data-access tiers, model-ownership rights and audit rules are not specified. It is also unclear how many international firms have joined or how many returned models are already deployed at the front.

The performance claims around Avengers, including weekly detections and interception automation, are attributed to Ukrainian reporting and ministry statements cited in the source material. Independent validation, false-positive rates and performance under heavy Russian electronic warfare have not been provided.

Returned Models Reach Drones

The next test is whether models trained inside the Dataroom move quickly into Ukrainian systems such as DELTA, VEZHA and front-line drones. The clearest milestones will be broader onboard machine vision, better operation under jamming and practical counter-Shahed automation with human authorization still in the loop.

Allied governments and defense companies will also be watching the access rules. If the model works, Ukraine could become a central training ground for battlefield AI while keeping control of the data that makes those systems useful.

Key Questions

What is Avengers Labs?

Avengers Labs is described as a Ministry of Defense partnership platform inside Ukraine’s Brave1 defense-innovation cluster. It lets selected defense companies train AI models on annotated combat-drone data in a protected environment.

Is Ukraine selling raw combat footage?

No. The source material says companies work inside the Brave1 Dataroom and do not take raw footage away. Ukraine keeps the data under control and receives the improved models.

Why is this dataset valuable?

It comes from real combat drone missions, including night, weather, camouflage and electronic-warfare conditions. Those examples are difficult to scrape from public sources or simulate accurately.

Are these autonomous weapons?

The program is aimed at detection, tracking, navigation and drone interception support. The source material says human operators remain responsible for firing decisions in the Shahed interception workflow.

What remains unconfirmed?

The public material does not give full access terms, pricing, independent performance testing or a list of foreign participants. Claims about detection volume and automation remain attributed to Ukrainian sources.

Source: Thorsten Meyer AI