Author: Krystyna Biletska / CEA
On September 16th 2025, CEA organised workshop AI4Cyber 2025 in Paris. It was dedicated to exploring the cutting-edge applications of Artificial Intelligence in strengthening cybersecurity and securing cyber-physical systems. The aim was to delve into novel AI-driven approaches for threat intelligence, proactive defense mechanisms, anomaly detection, and automated incident response.
Inspired by the research carried out in KINAITICS, which investigates AI-based behavioral monitoring and defense strategies against complex threats, this event showcased how AI can augment classical cybersecurity tools and address human factors and uncertainties in dynamic threat landscapes.
The event brought together 22 researchers and practitioners, including 2 key speakers and 5 speakers. Among 8 presentations, 4 scientific articles and one book were highlighted.
Research findings in the following areas were presented:
- Semi-automated penetration testing with LLMs & attack graphs
• Mimicry attacks that trick binary classifiers
• Multi-objective cyber response optimization with evolutionary algorithms
• Enhancing IDS with uncertainty quantification & explainability
• Explainable detection of process-oriented attacks in industrial systems
• Improving bot detection via retraining with counterfactual examples & active learning
The participants shared the common observation that moving beyond pure detection to trustworthy, interpretable, and resilient AI-driven security solutions is essential as cyber threats grow more sophisticated.
The program and several articles remain available on the webpage