UK offers up to 12 companies access to Ukraine's battlefield-data platform to build AI for drone swarms

The Ministry of Defence has opened the first competition under the UK–Ukraine AI Partnership, offering up to 12 UK companies access to Ukraine's Avengers AI Labs battlefield data to build AI for drone swarms, according to a GOV.UK release published on 25 September 2026. We covered the August deal that set up the partnership.

Who runs it

The release says the work is "being led by the MOD’s Taskforce RAID (Rapid AI Delivery)", working jointly with the Prime Minister's AI Taskforce. The competition document describes TF RAID as established by the Prime Minister and the Defence Secretary, reporting "directly to the Chief of the Defence Staff". The competition sits under TF RAID's "EXECUTE" pillar, described as "swarms of AI-enabled uncrewed systems".

The release says the UK is "the first international partner to access Ukraine’s coveted Avengers AI Labs". Ukraine's Ministry of Defence, in its own announcement the same day, also calls the UK "the first international partner of Ukraine whose defence companies will gain access".

The data

The sources describe the dataset in different units:

• The GOV.UK release: "sensor information from over 6 million detections of objects, such as tanks, artillery, air defence systems, infantry, and aerial targets, including Shahed drones and reconnaissance Uncrewed Aerial Vehicles."
• The MOD competition document: "5M+ frames of real-world data" and "Millions of annotated objects".
• Ukraine's Ministry of Defence: "an annotated dataset of more than 5 million frames", mostly from unmanned systems via its DELTA combat system, "anonymised and delayed so that it poses no risk to troops".

According to Ukraine's ministry, participants "will not get direct access to the dataset and cannot download it" and instead train their models "within the secure perimeter of Avengers Labs". It adds that companies will get access "once the Ukrainian side has completed export control procedures".

What companies must build

Proposals must address one or more of four areas, which the release lists as:

• Autonomous Target Recognition: "AI-enabled detection, classification, prioritisation and tracking of operationally relevant objects across complex and contested environments."
• Distributed Decision-Making: "Enabling multiple autonomous systems to dynamically adapt routes, priorities and behaviours in response to changing battlefield conditions, without relying on a single point of control."
• Adaptive Mission Execution: "AI approaches enabling multiple autonomous systems to coordinate actions under constrained communications and decentralised control."
• Collaborative Sensing and Information Fusion: "Methods for sharing, fusing and exploiting information across multiple platforms and sensors to improve situational awareness and mission effectiveness."

The release says the competition "could see the creation of models that let drone swarms keep working when communications are degraded or satellite navigation is unavailable". Proposals "must be applicable to both Ukrainian and British defence requirements".

Human control and safeguards

The release says drone swarms use AI to "identify threats and targets and reach decisions, always with appropriate human involvement". It says a small number of personnel "could direct large numbers of drones at once, on tasks from logistics resupply to precise targeting of enemy capabilities and hunting submarines in the North Atlantic." The release does not define "appropriate human involvement" and does not mention international law or an ethics review.

The competition document says TF RAID works "safely, securely, and with clear human accountability". It says the MOD will conduct due diligence on applicants, with "continued oversight of the approved companies accessing the Avengers Labs platform by the MOD and Ukraine", and asks applicants: "How will your organisation ensure the secure and responsible use of data?"

Timeline and money

The competition document sets out this timeline and two phases:

• Submissions close at 23:59 BST on 22 October 2026, with an industry briefing on 7 October.
• The MOD reviews bids from 23 to 30 October; selected companies are notified "by 06 November 2026" (its timeline says the week commencing 2 November).
• Phase 1 gives up to 12 companies free access "via a non-financial contract" for three months. The document gives this window as "approximately late November 2026 to late February 2027" in one place and "(Approx) mid November 2026 – mid February 2027" in another. Selected companies "shall bear all of the costs incurred".
• Phase 2, "from early 2027", awards "funded MOD contracts" to up to five companies. No sum is stated.

The document also says "Ukraine shall own the trained AI model weights developed by UK companies", with a licence to the MOD and a sublicence to the company.

What the ministers said

Defence Secretary, Wes Streeting MP said: "This world-first access to real battlefield data will allow British companies to continue pushing the frontier of defence innovation, deterring our enemies and making us safer at home as well as further strengthening Ukraine."

AI Minister Kanishka Narayan said: "This is the Prime Minister’s AI Taskforce delivering: bringing together Ukraine’s hard-won battlefield experience with Britain’s best AI companies to develop and deploy capabilities faster."

Background

GOV.UK's 24 August release says the partnership was signed by President Volodymyr Zelenskyy and Prime Minister Andy Burnham in Kyiv, and that pilot projects had been rolled out with three British startups: Sintela, Mind Foundry and Skyral.

Why it matters

On our reading, the competition document shows what the August deal means for companies: Phase 1 is a non-financial contract in which they bear their own costs and, according to Ukraine's ministry, cannot download the data, while Ukraine keeps ownership of the model weights they train. The release describes human involvement in one phrase and does not set out rules for the "precise targeting of enemy capabilities" it describes.