Integrating Artificial Intelligence in the European Union’s Civilian Common Security and Defence Policy Missions

The authors find a number of potential benefits to AI use that have applicability throughout civilian CSDP. AI's data analysis potential means that vast quantities of data can be imported and analysed with great speed. And AI-powered pattern recognition can both find gaps in current knowledge and create tailored outputs for different people with different needs. 

However, there are significant caveats due to technical and human limitations. For technical limitations, AI's potential for hallucination and sycophancy mean that there is a risk of "garbage in, garbage out" where flawed input data can produce equally flawed outputs hidden behind professional-looking presentation. This is especially a risk if the input data is poisoned by hostile actors. In terms of data processing, the reasoning models of AI models is opaque to the user, meaning if output errors are found their exact cause can be difficult to ascertain. Furthermore, regardless of technical risks, ensuring AI literacy among all personnel using this new technology is of key importance and mandated to providers and deployers of AI by the EU AI Act.

The following two sections of the paper explain how all these potential benefits and risks manifest first in training for civilian CSDP and then in the planning, implementation and evaluation of missions. For example, AI could assist in identifying training needs from vast quantities of data contained in lessons learnt, after-action reviews and other forms of student and instructor feedback. However, if important issues to address are underreported or underrepresented in data the AI will deemphasise these topics in favour of what the data shows. And in the field, AI can assist in early warning systems by monitoring news and social media, but it is still at risk of data poisoning if malign actors create input data with the intent to mislead the AI's analysis.

The authors conclude that given both the strength of potential usefulness and severity of the risks, AI integration should be carried out cautiously, beginning in a limited way, with strong safeguards:

• Integrate AI initiatives within existing policies and processes, using existing EU and international legal frameworks in the absence of official civilian CSDP AI policy.
• Analyse the potential impacts of each specific AI use case rather than a general or abstract analysis.
• Ensure AI literacy and privilege human judgment in all areas.
• Procurement of AI tools should ensure EU digital sovereignty to avoid data risks and foreign dependency, and meet sustainability requirements.
• Integration of AI should begin with progressive, limited, documented, and reversible pilot programs to assess how risks manifest in practice.

Reference: Levantino, F. and Stroppa, M. (2026). Integrating Artificial Intelligence in the European Union’s Civilian Common Security and Defence Policy Missions. European Union Training Initiative (EUTI); Sant’Ana Scuola Universitaria Superiore Pisa.