A student's dream: hacking (then fixing) Gradescope's autograder - Aditya's Blog
Ever since I started exploring security more deeply, I’ve been asked countless times by people if I could hack into grading systems to change my (or, more often, their) grades. With Gradescope being the most ubiquitous platform for grading STEM classes at Stanford, my standard response was always that I couldn’t, imagining that a well-established EdTech company would secure their platform well enough. As it turns out, Gradescope’s autograders have been vulnerable to various types of attack since 2016. Gradescope has known about the issues since at least 2020, yet has indicated it cannot distribute a general fix. This post covers my exploration of Gradescope’s autograder vulnerabilities, an analysis of the potential impact on courses, and how I created Securescope, my attempt at a more secure autograder configuration.
February 28, 2023 Ever since I started exploring security more deeply, I’ve been asked countless times by people if I could hack into grading systems to change my (or, more often, their) grades. With Gradescope being the most ubiquitous platform for grading STEM classes at Stanford, my standard response was always that I couldn’t, imagining that a well-established EdTech company would secure their platform well enough. As it turns out, Gradescope’s autograders have been vulnerable to various types of attack since 2016. Gradescope has known about the issues since at least 2020, yet has…
saved by
related reading
- Gradescopegradescope.com
- Training a Misaligned Reward Seekeralignment.anthropic.com
- Security incident disclosure — July 2026huggingface.co
- [2312.06942] AI Control: Improving Safety Despite Intentional Subversionarxiv.org
- Vulnerability Research Is Cooked - Quarrelsomesockpuppet.org
- Investigating three real-world incidents in our cybersecurity evaluations \ Anthropicanthropic.com
- A basic systems architecture for AI agents that do autonomous research — LessWronglesswrong.com
- Most Neoclouds Suck At Securitynewsletter.semianalysis.com
- Hacking OpenAIhacktron.ai
- GitHub RCE Vulnerability: CVE-2026-3854 Breakdown | Wiz Blogwiz.io
- Online Judge / LeetCode System Design | System Design Interviewsystemdesignschool.io
- Recent Frontier Models Are Reward Hacking - METRmetr.org