CDUP: Detecting AI Generated Code Using Abstract Syntax Tree Analyst

Recent generative AI development, especially code generation assistants, prompts the dilemma of how to identify plagiarised code among students’ submissions. This problem is identified as an increasingly important one, with reports claiming an increase of plagiarism ranging from 15 to 45% in various areas. Most work in the field is aimed at the larger problem of detecting AI generated text, leaving an important gap as far as specialized tools for code detection are concerned. Techniques such as stylometry [1], [2], metadata analysis [3] or probability curvature analysis [4] are significant, yet they do not exploit the structure, syntactic or semantic properties of source code. In this paper we aim to inspect the Abstract Syntax Tree (AST) of a targeted piece of code and use clone detection techniques to identify potentially AI generated code. The AST is a very rich data-structure that maintains considerable information from the source code. Therefore, it is suitable for various types of analysis and offers significant flexibility. The tool is applied over small, yet relevant, samples of programs written in C, offering important information about the limitations of our analysis as well as indicators for future work.