Work

Tekoälyn tietoturva : Tekoälyn yleisimmät riskit, riskienhallinta ja EU:n tekoälyasetus

Year: 2025
Type: dissertation
Abstract: The topic of this literature review is the security of artificial intelligence (AI). As AI becomes more common and its use expands, the topic is both appropriate and of general interest. As the EU Artificial Intelligence Act (AI act) is now in force, the regulation of AI and its impact on AI systems, their development, and security is becoming increasingly highlighted. The review focuses on the risks related to AI and their management. The objective is to identify the most common risks and to find out which control measures are suitable for managing them. This point of view is examined through the research question: What are the most common cybersecurity risks of AI, and how can they be managed? In addition, the review studies the EU AI Act and its implications for the development and use of AI systems within the European Union. The related research question is: How does the EU AI Act affect the cybersecurity practices and requirements of AI systems within the EU? The analysis method used in the review was content analysis, which focused on finding the meanings and themes of the texts in the material and categorizing them. In the review of the most common AI risks and their management methods, the primary source was the ProQuest database, and the secondary source was the Google Scholar search engine. The data collection was narrowed to the period from January 1, 2020, to April 16, 2025. Simple keywords were used in the search process without search phrases. The search results were evaluated individually and assessed for their relevance to the research question. A total of 31 peer-reviewed publications were chosen for the final analysis. In the analysis, all mentioned risks and their management methods were collected from the research material. The gathered risks were grouped, and the frequency of occurrence for each risk group was calculated from the material. Based on this, the most common risk groups were identified. For the six most common risk groups, control measures were searched from the research material. The primary source for analyzing the requirements of the AI Act was the EU AI Act itself, and the secondary source was the guidelines and publications issued by the European Commission. The regulation and other materials were reviewed, and the relevant sections were organized and analyzed in relation to the research question. In addition, two case analyses were conducted on hypothetical AI systems. These case studies aimed to demonstrate the effects of the regulation on AI systems. Based on the analysis, the most common AI risks included risks related to model and data manipulation, privacy and data protection, vulnerabilities in AI systems, model bias, the use of AI in cyberattacks, and ethical issues related to AI use. The requirements of the EU AI Act try to ensure the safety and reliability of AI systems and to protect fundamental rights. The regulation prohibits some uses of AI. Prohibited systems include those that manipulate or mislead users, exploit user vulnerabilities, conduct unfair social scoring, or enable large-scale surveillance without a legal basis. For high-risk AI systems, the regulation requires, for example, thorough documentation and continuous monitoring throughout lifecycle, registration in the EU database, CE certificate, demonstration of compliance, accuracy, stability, and cybersecurity appropriate to the intended use case through the system's lifecycle, as well as human oversight. For low-risk systems, the regulation sets transparency obligations. Users must be told when interacting with AI. Synthetic products created by AI must be labeled as artificially generated.
Author: Tölli, Mari
Language: Finnish
Subfield: Law
Open Access status: green
OpenAlex ID: W7133075199
Cites: 34
Cited by:
Related to: 10