Answers are grounded in this paper and cite their sources.
Software security is a pressing concern for the industry, yet Computer Science (CS) graduates often lack practical skills in secure programming. At the same time, Large Language Models (LLMs), such as ChatGPT, are rapidly being adopted by developers, including for security-related tasks, though their unregulated use raises concerns in practice. In this paper, we report on the design, enactment, and reflection of a teaching plan for undergraduate students in CS that integrates industrial expertise, academic instruction, and the use of ChatGPT to fix vulnerabilities in e-commerce web apps. We involved 105 students and combined lessons delivered by security practitioners and academics with laboratory and homework assignments in which students fixed real vulnerabilities with or without the aid of ChatGPT. The experience gained from the enactment of our teaching plan shows that students using ChatGPT were able to fix significantly more vulnerabilities, but also introduced more incorrect fixes, highlighting both opportunities and risks. When supported by ChatGPT, the overall process for fixing vulnerabilities was perceived as easier. Moreover, most students valued the experience, reporting improved knowledge on securing web apps. Overall, this paper contributes an educational experience report that illustrates how LLMs can be integrated into the teaching of secure web development and discusses implications on the integration of ChatGPT into teaching secure web development. For example, a lecturer should make clear that ChatGPT provides a valuable support for vulnerability fixing, even if it is not a silver bullet.
No takes yet. Share an insight, caveat, or question.
Nocera et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: