EU AI Act Rules for Defence Contractors’ Dual-Use AI Systems

Dual-Use AI Systems and the Rules That Still Apply

The EU AI Act excludes AI systems developed and used exclusively for military, defence or national security purposes from its scope entirely, under Article 2(3). Systems with a dual-use or civilian application, including AI used in general business operations at defence contractors, are not covered by that exclusion, and for those systems the same deferral logic applies as everywhere else: Annex III high-risk obligations move to December 2027, but prohibitions, general-purpose model obligations, and transparency duties were never deferred.

The practical risk for a contractor is assuming that because military use is excluded, every AI system it touches is exempt. That assumption breaks down the moment a system has any civilian or dual-use application, at which point the full, largely undeferred Act applies. This article sets out where the line sits and what a contractor’s in-scope systems actually owe the Act today.

Executive Summary

Defence contractors need to draw a clear line between AI genuinely exclusive to military or national security use, which sits outside the Act entirely, and dual-use or general business AI, which remains fully subject to obligations that were never deferred, including transparency and general-purpose model duties. For programme security and compliance leads, the practical task is classifying each AI system against that line before assuming the deferral, or the exclusion, covers it.

Key Takeaways

1. Purely military-use AI sits outside the AI Act’s scope by design, under Article 2(3), and none of the deferral discussion applies to it because it was never in scope.
2. Dual-use and non-defence-specific AI systems used by defence contractors, such as general business tools, remain fully in scope, deferral included.
3. Only Chapter III high-risk obligations for in-scope systems were deferred, not the whole Act.
4. Transparency obligations under Article 50 were not delayed for any in-scope system and are already active.
5. Prohibited AI practices remain banned for any in-scope system regardless of the high-risk timeline.

What the Military Exclusion Actually Covers

Article 2(3) removes AI systems developed and used exclusively for military, defence or national security purposes from the Act’s scope. This is a narrower exclusion than it sounds.

Exclusivity Is the Operative Word

A system built and used only for a military or national security purpose is excluded. A system also sold or deployed for civilian purposes, or used for general business functions like HR or finance within a defence contractor, is not covered by the exclusion simply because the contractor operates in the defence sector.

Dual-Use Systems Fall Back Into Full Scope

Once a system has any non-military application, it is evaluated under the Act on the same terms as any other AI system, including the deferral timeline that applies to genuinely in-scope high-risk systems.

What the Digital Omnibus Actually Deferred for In-Scope Systems

For AI systems that are in scope, the deferral mechanism covers Chapter III high-risk obligations, currently pushed to December 2027, because the technical standards needed to operationalise them were not ready in time.

A Targeted Delay to Chapter III, Not a Reset of the Act

Everything outside Chapter III, prohibitions, general-purpose model obligations, transparency duties, proceeds on the timeline the Act originally set for any in-scope system.

What Is Already in Force for In-Scope Systems, Deferral or Not

Transparency and disclosure duties, and obligations on general-purpose AI model providers, applied on the Act’s original schedule and were not moved by the Omnibus for any system not excluded under Article 2(3).

Building AI Governance Around the Scope Line, Not an Assumption

The practical starting point for a contractor is classifying each AI system against the military exclusion first, then applying the standard obligation calendar to everything that falls outside it.

How a Data Control Plane Supports AI Governance Obligations Already in Force

The Kiteworks Data Control Plane applies data-aware, zero-trust controls to every action any user or AI agent takes across every channel, and captures each one in a tamper-proof, unthrottled audit log that feeds directly into SIEM tooling, giving contractors the detailed, queryable record current transparency obligations expect for their in-scope AI systems.

Contractors that want to see how this applies to their in-scope AI deployments can schedule a custom demo.