Ensurance of artificial intelligence systems cyber security: analysis of vulnerabilities, attacks and countermeasures

Abstract

In recent years, many companies have begun to integrate artificial intelligence systems (AIS) into their infrastructures. AIS is used in sensitive areas of society, such as the judicial system, critical infrastructure, video surveillance, and others. This determines the need for a reliable assessment and guaranteed provision of cyber security of AIS. The study analyzed the state of affairs regarding the cyber security of these systems. Possible types of attacks are classified and the main ones are considered in detail. Threats and attacks were analyzed by level of severity and security risks were assessed using the IMECA method. “Adversarial attacks” and “Data poisoning” attacks are found to have the highest risks of danger, but the countermeasures are not at the appropriate level. It was concluded that there is a need for formalization and standardization of the life cycle of the development and use of secure AIS. The directions of further research regarding the need to develop methods for evaluating and ensuring cyber security of the AIS are substantiated, including for systems that provide AI as a service.