Uncomfortable questions on regulating AI as a dual-use technology
Artificial intelligence is a textbook case of dual-use technology, both civilian and military. The AI Act expressly excludes military uses, but the line with national security, intelligence and autonomous defence systems raises uncomfortable questions about democratic oversight, exports and liability.
AI as an inherently dual technology
Artificial intelligence is the clearest example of dual-use technology: the same computer-vision algorithm can diagnose a tumour, assist a self-driving car or guide an armed drone. The same language model can help draft a contract or automate disinformation campaigns.
European regulation has opted for a formal separation: the AI Act (EU Regulation 2024/1689) excludes from its scope AI systems developed or used exclusively for military, defence or national security purposes (Art. 2(3)). The EU Regulation 2021/821 on dual-use items governs their exports. Between them sits a grey zone raising uncomfortable questions.
> Practical note: tech companies serving both civilian and public customers (administrations, law enforcement, defence) already operate in that grey zone. Due diligence is unavoidable.
Uncomfortable questions the regulation leaves open
1. Where does "civilian" end and "military" begin?
A facial recognition system sold to a local council may end up being used by intelligence services. A commercial civilian drone can be weaponised in hours. Classification depends not on the product but on the end user.
2. Who oversees military AI systems in the EU?
The AI Act does not apply to them. There is no European equivalent. Each Member State keeps its own mechanisms, with limited transparency and very uneven parliamentary control.
3. Is "responsible use" really possible in autonomous weaponry?
Lethal Autonomous Weapon Systems (LAWS) have been debated at the UN for over a decade. The EU still lacks a binding common position. The central question — can a machine decide to kill? — remains regulatorily unanswered.