DATASETS UNVEILED: CHALLENGES IN TRAINING AI FOR THE SECURITY DOMAIN
No Thumbnail Available
Date
2025-12-15
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Plovdiv University Press
Abstract
The rapid evolution of technology, and the recently adopted legal frame-work at the European Union level, poses significant challenges for both deployers and providers of AI-based systems. Ensuring the quality of datasets used to train artificial intelligence is crucial for meeting the requirements of robustness, transparency, and accuracy. It is essential to verify the origin and method of collection or creation of these datasets, as well as the mechanisms though which they are used to train AI models. These steps are mandatory to ensure the ethical and legally compliant development of AI systems. In the security domain, AI-based systems are increasingly employed by law enforcement agencies for operational purposes and the maintenance of public order. These systems are often classified as high-risk due to their substantial impact on individual privacy, particularly through profiling and the use of big data (point 6, Annex III to Regulation (EU) 2024/1689). Therefore, it is imperative to ensure data protection by design and to establish appropriate legal requirements for the development of such software.
Description
Keywords
artificial intelligence, datasets, security