Abstract <p>The article is dedicated to the study of intelligent fuzzing methods aimed at improving the security of aviation systems software using machine learning algorithms. Software used to control aviation systems, such as on-board networks and air traffic control systems, requires special attention due to the high risks of unauthorized tampering. Analysis of conventional fuzzing methods has revealed their disadvantages, such as redundancy of input data and significant computational costs, which limit the overall performance and effectiveness of testing aviation information systems within the secure development cycle. The authors propose to solve these problems by using machine learning to predict the results of program execution and generate test inputs that are more likely to identify critical vulnerabilities. This approach reduces redundant iterations and increases code coverage, which helps to identify errors at earlier stages of development of aviation systems software. The article presents the results of experiments that confirmed the high effectiveness of the proposed method based on AK-VS 3 ModFuzz, which demonstrates superiority over traditional tools such as AFL, both in terms of the number of new program execution paths discovered and in terms of testing time. This approach results in a more efficient testing process, which contributes to improving the security level of aviation systems software and reducing the time spent on its development.</p>