Managing criminal-law risks in the use of artificial intelligence: The framework of the Council of Europe and the Republic of Serbia, with reference to European Union law

The increasing use of artificial intelligence systems raises important questions regarding criminal liability and the management of legal risks associated with their application. A particular challenge is determining how existing criminal law concepts apply when AI systems are used in ways that may result in harm to legally protected interests, and whether current normative frameworks are adequately equipped to address such situations. The aim of this paper is to analyse the criminal law framework relevant to liability for the use of artificial intelligence in the Republic of Serbia, with particular emphasis on Council of Europe instruments and a supplementary overview of relevant European Union law. The paper examines the ways in which existing criminal law concepts can be applied in situations where artificial intelligence is used as a means of committing criminal offences or as part of a complex chain of decision-making leading to a prohibited consequence. The research is based on normative and comparative methods and includes an analysis of relevant institutional and legal sources, including the CDPC Feasibility Study on Artificial Intelligence and Criminal Law (2020) and the Council of Europe Framework Convention on Artificial Intelligence (2024). The analysis shows that the comparison must be grounded primarily in criminal-law instruments rather than in regulatory frameworks with different purposes. At the level of the Council of Europe, and with supplementary reference to relevant European Union regulatory law, the examined instruments provide standards concerning human oversight, transparency and accountability that may serve as normative parameters for assessing due diligence, negligence and culpability in concrete cases. In the Republic of Serbia, the absence of such specific standards creates greater legal uncertainty in complex cases involving autonomous systems and multiple actors in the chain of responsibility. All examined frameworks share the foundational principle that artificial intelligence cannot be a subject of criminal liability; responsibility remains with the human actors and, where the statutory conditions are met, legal persons involved in decisions about the development, deployment and supervision of AI systems. The results indicate that the effective management of criminal-law risks associated with the use of artificial intelligence requires well-defined normative standards, organisational control mechanisms and explicit responsibilities of entities involved in the development and use of such systems. The further development of Serbia's normative framework, drawing on Council of Europe instruments relevant to the criminal-law assessment of artificial intelligence, can contribute to more effective risk management and better alignment between technological development and criminal-law protection of social values.