Contemporary Approaches for Securing Autograders

This survey examines isolation techniques for securing autograders against untrusted student code in educational environments. Amid the increasing adoption of online autograders for evaluating submissions (e.g., Jupyter Notebook), robust security is essential to mitigate risks including cheating, unauthorized access, and malicious exploits. We review code isolation methods system call interposition, kernel primitives, and virtualization and evaluate their efficacy in educational contexts based on ease of deployment, exploit resistance, performance overhead, and educator usability. Analysis of previous research reveals strengths in kernel - and container-based solutions for scalability and efficiency, while virtual machines provide superior isolation for high-risk scenarios. Limitations, such as kernel vulnerabilities in shared environments, are addressed alongside emerging trends such as hybrid models and AI-assisted filtering. This work provides practical guidelines for educators to implement secure autograders, ensuring fair assessment and institutional resource preservation.