Flanders - Impuls program Defence: AI and Autonomous technology

The Flemish Public Service for Employment, Economy, Science, Innovation and Social Economy (WEWIS) has commissioned the development of two exploratory roadmaps as part of the Flemish Innovation and Industry Strategy for Security and Defence (VISD), approved by the Flemish Government on 19 December 2025. B8 – Exploratory Roadmap on AI for Defence (imec, October 2025) identifies 227 companies at the intersection of AI and defence and defines four focus areas comprising 14 building blocks. B10 – Exploratory Roadmap for Autonomous Technology for Security and Defence (Flanders Make, October 2025) identifies more than 120 companies and structures two pillars with eight building blocks around fleet innovation and autonomous missions.

Both roadmaps are based on a reverse dual-use logic: civilian AI and autonomous technology are explicitly channelled towards military applications, and not the other way round (which is usually the standard approach). This creates a concrete innovation agenda for Flemish industry, but both documents remain silent on the ethical and governance implications of this shift, an observation that deserves attention in the context of the far-reaching military exemption in the EU AI Act. This analysis will briefly discuss all three documents.

The two roadmaps are products of the Flemish Innovation and Industrial Strategy for Security and Defence (VISD), the overarching Flemish strategy for innovation in defence and security. The strategy’s vision is for Flanders to grow into an internationally leading region within the European security and defence value chain, both as a frontrunner and as a partner.

The VISD is structured around three strategic objectives: (1) strengthening Flemish industrial defence capabilities, (2) increasing the innovation capacity of the Flemish security and defence ecosystem, and (3) stimulating economic growth through defence-related industry and R&D. Eight strategic domains are identified as priorities: Maritime, Aviation/(C-)UAS, Space, AI, Cybersecurity, Biotechnology, Autonomous Systems, and Energy/Environment/Geo-intelligence.

Implementation follows a layered model: the VISD provides the strategic framework; domain-specific roadmaps (B1–B10) define the innovation agenda; projects are generated from these; and supporting instruments – including WEWIS, VLAIO, FIT, PMV, LRM, FWO, and the Flemish Defence Fund – provide funding and facilitation. Shared O&I infrastructure, communication initiatives such as the Defence Innovation Flanders Days (DIF-Days), and monitoring via KPIs complete the implementation model.

A defining conceptual choice in both the VISD and the two roadmaps is the explicit reversal of the traditional dual-use direction. Historically, defence R&D drove civilian innovation: GPS, the internet, and radar technology all originated in military programs. The VISD and B8 argue that this logic has reversed: civilian AI is now the primary driver of defence applications, not the other way around. The Flemish commercial AI ecosystem – SMEs, start-ups, and research institutions – is the starting point. The VISD formalizes this through a “push-pull” approach: Flemish civilian technological strengths are “pushed” toward identified defence capability gaps.

This civilian-to-military direction is governance-relevant. It means that technologies developed under civilian regulatory frameworks, including the EU AI Act, are deliberately channeled into military applications that are exempt from that same framework. The strategy documents do not address how this transition is governed, what oversight applies at the point of conversion, or what safeguards travel with the technology from civilian origin to military application.

The B8 roadmap, developed by imec, maps 227 Flemish companies (270 including Brussels) at the intersection of AI and security/defence. A key finding is that more than half are not yet publicly active in the defence domain, indicating substantial latent potential according to the strategy. The roadmap translates identified defence needs (at national, European, and NATO levels) into four focus domains:

• Focus 1 – AI Sensing: four building blocks covering optical detection, distance-based detection, time-series analysis, and sensor fusion.
• Focus 2 – AI Compute: three building blocks covering edge computing, new hardware paradigms, and compute/test infrastructure & sandboxing.
• Focus 3 – AI for Communication: three building blocks covering secure communication, identification/tracking/jamming, and positioning/navigation/timing.
• Focus 4 – AI for Operational Support: four building blocks covering logistics and decision support, human-AI interaction, simulations/training, and data & ICT infrastructure.

A cross-cutting theme across all four domains is software-hardware co-design, explicitly positioned as both a performance requirement and a requirement for digital sovereignty. The roadmap identifies Focus 4 as the domain where the Flemish ecosystem aligns most broadly, while Focus 1–3 require more specialized contributions from a smaller group of actors.

The roadmap also proposes specific testing environments for defence AI (regulatory sandboxes) where temporary regulatory flexibility would apply.

4. B10 – Autonomous Technology: 2 pillars and 8 building blocks

The B10 roadmap, developed by Flanders Make, maps more than 120 companies active at the intersection of autonomous technology and security/defence, with the vast majority currently focused on civilian markets. The roadmap is structured around two pillars:

• Pillar 1 – Strengthening the innovation cycle at fleet level (3 building blocks): Digitized end-to-end lifecycle management (BB1.1), distributed microfactories (BB1.2), and production techniques & ReX methods (BB1.3). These building blocks cover the full lifecycle of autonomous platforms, from development to deployment and maintenance.
• Pillar 2 – Resilient autonomous missions with hybrid fleets of drones/UXVs (5 building blocks): The roadmap aligns with the NATO/EDA taxonomy for autonomous systems and distinguishes three levels of human control: human-in-the-loop (HITL), human-on-the-loop (HOTL), and human-out-of-the-loop (HOOTL). Notably, the “Decision-making” building block (BB2.3) retains human-on-the-loop as the standard, while the other execution-oriented building blocks are framed as human-out-of-the-loop – a distinction that is operational rather than ethical.

The VISD and its accompanying roadmaps outline a clear implementation trajectory. In the short term, the priority is translating the building blocks into concrete, fundable projects that can be submitted to Belgian, European, and NATO funding mechanisms (including the European Defence Fund and EDIP). The strategy foresees the development of flagship projects demonstrating Flemish leadership in priority domains.

In parallel, efforts focus on ecosystem building: the Defence Innovation Flanders Days (DIF-Days) are intended to become a recurring platform for matchmaking between government, industry, and knowledge institutions. WEWIS acts as programme coordinator, while VLAIO, FIT, PMV, LRM, and FWO provide supporting instruments (grants, loans, risk capital, research funding).

The strategy and the orientation roadmaps do not contain an explicit ethical framework, red lines, categories of prohibited use, or a do’s-and-don’ts list for military AI or autonomous systems. The VISD strategy does mention an “Ethical and Strategic Advisory Committee” (ESAC) in its list of abbreviations, but the mandate, composition, powers, and advisory processes of this body are not elaborated anywhere in the strategy documents. Whether the ESAC is operational, what it advises on, and how its recommendations feed into programme and project decisions remains unclear.

The building blocks and ecosystem instruments provide concrete access to European and NATO funding, but the transition from civilian to defence markets brings regulatory and commercial risks. Export controls under the EU Dual-Use Regulation (2021/821) impose licensing requirements on software and technologies with both civilian and military applications, creating a significant compliance burden for companies without specialized legal infrastructure. Liability exposure is a related concern: if a civilian AI product adapted for military use causes harm, responsibilities between developer, integrator, and end client are legally unclear without a clear governance framework.

Re-entry into civilian markets presents a structural challenge for companies operating across both domains. A product refined under a defence contract (trained on operational data and optimized for threat detection) will not automatically meet conformity assessment requirements under the EU AI Act when redeployed in civilian applications. Reputational and funding risks add another layer: ESG-sensitive investors and some EU funding instruments impose restrictions on military applications, and visible defence ties can create friction in civilian public procurement.