Friday 28 March 2025
The EU’s Artificial Intelligence Act has recently been adopted, aiming to create a legal framework for the development and deployment of trustworthy artificial intelligence (AI) systems. As part of this effort, researchers are working to clarify the legal terminology surrounding AI, particularly in regards to robustness and cybersecurity.
One key aspect of the AIA is its classification of AI systems into different risk categories. High-risk AI systems, or HRAIS, are those intended for use in critical infrastructure, such as transportation or healthcare, where a failure could have significant consequences. These systems must meet strict requirements for robustness and cybersecurity.
Robustness refers to an AI system’s ability to maintain its performance even when faced with unexpected inputs or changes in the environment. In other words, it can adapt and continue functioning correctly, even if something goes wrong. Cybersecurity, on the other hand, is about protecting against unauthorized access, data breaches, and other forms of digital attack.
The AIA requires providers of HRAIS to ensure both robustness and cybersecurity throughout the system’s lifecycle – from development to decommissioning. This means that developers must consider not only how their AI systems will perform during normal operation but also how they will respond to unexpected events or attacks.
One challenge in achieving this balance is the trade-off between accuracy and robustness. Machine learning (ML) models, which are often used in AI systems, can be highly accurate but may also be vulnerable to certain types of attack. For example, an attacker might try to trick an ML model into misclassifying data or generating false outputs.
To address this issue, developers must carefully consider the design and implementation of their AI systems. This includes selecting appropriate algorithms, training data, and evaluation metrics to ensure that the system is both accurate and robust. It also means being transparent about any trade-offs made during development and documenting these decisions in technical documentation.
Standards for HRAIS will play a crucial role in ensuring compliance with the AIA’s requirements. These standards should provide clear guidelines on how to achieve the necessary balance between accuracy, robustness, and cybersecurity. They may also offer guidance on specific metrics and processes for evaluating AI system performance.
The development of these standards is an ongoing process, and researchers are working to ensure that they take into account the latest advances in ML and AI. By providing a clear framework for the development and deployment of trustworthy AI systems, the EU’s AIA aims to promote innovation while minimizing the risks associated with AI technology.
Cite this article: “EUs Artificial Intelligence Act: Ensuring Trustworthy AI Systems through Robustness and Cybersecurity”, The Science Archive, 2025.
Artificial Intelligence, Eu, Robustness, Cybersecurity, High-Risk Ai Systems, Machine Learning, Accuracy, Transparency, Standards, Trustworthy Ai
