Ensuring Compliance of High-Risk AI Systems with the EU AI Act Using Open Source Libraries
Anja Obradov, Bogdan Pavković · 2025
The EU AI Act introduces new regulatory requirements for providers of AI systems, including high-risk AI systems, such as those used in the automotive and medical field. The regulations for the development and deployment of AI systems make compliance a critical challenge for AI developers and organizations, while promoting human-centric, trustworthy AI and mitigating the risks of harmful AI impacts. This paper analyzes the relevant chapters of the AI Act, translating legal obligations into technical requirements that AI developers must fulfill. We propose a compliance verification framework that integrates multiple open-source libraries to analyze key areas of the AI system, such as dataset quality, human oversight, accuracy, robustness, and transparency. In addition, we highlight the importance of benchmarking AI software using proposed methods to ensure regulatory compliance. The framework enables AI developers to proactively meet legal requirements, increasing the safety, reliability, and trustworthiness of AI systems.