RISKS AND LIMITATIONS OF USING NEURAL NETWORK TOOLS IN ACCOUNTING INFORMATION SYSTEMS

Objective: to investigate the risks and limitations of using neural network models in accounting information systems. Objectives: to form a classification and description of the limitations and risks of using generative models in the context of the specifics of management and accounting information systems, as well as to form final recommendations for reducing the risk burden when integrating neural network tools into accounting information systems. Methods: systematic review of the literature (key databases: Semantic Scholar, E-library.ru, Springer), qualitative analysis of the content of foreign and domestic publications, meta-analysis of secondary data in order to identify individual risks, comparative analysis of risk definitions. Results: risks and limitations were divided into two groups according to the stages of integration of neural network tools: the stage of generative model training and the stage of subsequent operation. The first stage includes restrictions related to data confidentiality, data markup and volume, accounting specifics, etc. The second stage was divided into 6 groups according to the origin and sphere of influence of these risks: technical, organizational, legal, informational, economic, reputational. A description of the limitations and possible ways to manage the negative consequences arising from them has been compiled. Recommendations for improving the control output of neural network tools are proposed. Conclusions: a risk-based approach to the problem of use will allow an economic entity to make a more informed decision regarding the integration of neural network tools into the accounting information system at the current stage of development of artificial intelligence technologies. A possible way to solve the problems described in the article has been identified — the multi-agent structure of the model, which is to be implemented in the accounting system, but this issue goes beyond the scope of the article and requires further research.