On Wednesday March 13, the European Parliament adopted the text of the AI Act, the first comprehensive binding regulation governing AI with 523 votes in favour, 46 against and 49 abstentions, setting a global precedent and affirming Europe’s leadership in the regulation of artificial intelligence.
AI Act: What impact for defence?
At Sierra Tango, we have carefully reviewed the measures set out in the text, the comments made by European institutions and the reactions of experts, in order to understand the implications of AI in the defence sector.
Risk-based approach
But first of all, a little context. This initiative proposed in 2021 by the Commission sets out guidelines for the introduction, deployment and use of artificial intelligence systems within the Union. It classifies AI systems according to the level of risk they are likely to generate and imposes different requirements accordingly.
High-risk systems must comply with strict standards to guarantee the safety of individuals and their fundamental rights. Cybersecurity protocols, human monitoring and risk mitigation measures are therefore required. Minimal-risk systems, on the other hand, have less stringent transparency obligations.
The law prohibits certain harmful uses of AI, such as real-time biometric identification in public spaces, biometric categorisation based on personal traits such as race or political opinions, individual predictive policing, and the untargeted collection of facial images.
Companies breaching these provisions will be subject to fines, up to 35 million euros or 7% of worldwide annual sales for serious breaches, 15 million euros or 3% for other breaches.
Broad exemption for national security
Article 2 §3 of the AI Act exempts from its application “AI systems that are intended exclusively for military, defence or national security purposes, regardless of the type of entity carrying out these activities”, including the companies responsible for developing these systems or external subcontractors.
In accordance with Article 4, §2 of the TEU and the specific features of the defence policy of the Member States and of the EU referred to in Chapter 2 of Title V of the TEU, the Act does not apply to areas not covered by EU legislation, and national security remains the sole responsibility of the Member States.
Ambiguity regarding dual-use application impacts
Despite this exemption, the application of the AI Act to AI systems could still bring additional complications and challenges for defence companies and military organisations.
Indeed, as mentioned by Rosanna Fanni of CEPS, the key to interpreting this exclusion from Article 2 lies in the term “exclusively”, which therefore suggests that AI systems used for dual purposes may still fall within the scope of the AI Act.
Should an AI system, initially developed and used for military, defence or national security purposes, be used for other purposes (civilian or humanitarian) on a temporary or permanent basis, it would then be subject to the regulations of the AI Act, obliging them to comply with the requirements and risk classification attributed to that system, and the consequences thereof.
Industrial sector interest and concerns
The level of interest is clear. In 2021, more than 50 partners, including Airbus, Naval Group, Safran, Thales and Sopra Steria, entered the French Confiance.ai programme to examine the issue.
This initiative reflects a desire to encourage the responsible development of artificial intelligence in Europe. Today, they are committed to facilitating the implementation of the Al Act by the industry.
However, concerns remain. It was expressed by 150 companies (including Dassault, Airbus and Safran) in an open letter published in the Financial Times in June 2023.
The signatories pointed to the potential threat posed by the Al Act to Europe’s competitiveness and technological sovereignty.
In conclusion, although the adoption of the AI Act by the European Parliament represents a significant step forward in the regulation of artificial intelligence, concerns over its application in the defence field remain.
The exemption granted to AI systems for military or national security purposes still raises a number of questions, particularly with regard to dual-use applications.
The vote in the European Parliament marks the second-to-last step before the AI Act is ratified by the EU Council, which is expected to vote in the coming weeks.