The development of Artificial Intelligence (AI) technology provides strong support for source code processing based AI systems. Compared with natural language processing, source code is special in semantic space. Machine learning tasks related to source code processing usually use the program’s abstract syntax trees, data dependency graph, control flow graph, and other methods to obtain the structured information of code and extract features. Existing studies can obtain excellent results in experimental scenarios through in-depth analysis of source code structure and flexible application of classifiers. However, for real application scenarios where the source code structure is more complex, most of the AI systems related to source code processing have poor performance and are difficult to implement in the industry, which triggers practitioners to consider the robustness of AI systems. Because AI-based systems are generally data-driven black box systems, it is difficult to directly measure the robustness of such software systems. With the rise of adversarial attack techniques, some scholars in the field of natural language processing have designed adversarial attacks for different tasks to verify the robustness of models and conducted large-scale empirical studies. To solve the problem of instability of AI systems based on source code processing in complex code scenarios, this paper proposes a Robustness Verification method (RVMHM, Robustness Verification by Metropolis-Hastings attack). Firstly, the code preprocessing tool based on the program’s abstract syntax trees is used to extract the variable pool of the model, and then the MHM source code attack algorithm is used to replace the variable perturbation model prediction effect. The robustness of the AI system is measured by observing the change in the robustness verification index before and after the attack by interfering with the data and model interaction process. This paper takes vulnerability prediction as a typical scenario of source code processing binary classification as an example and verifies the robustness of 12 groups of AI vulnerability prediction models on three datasets of open source projects to illustrate the effectiveness of the RVMHM method for robustness verification of source code processing based AI systems.