Workshops Archive

A First Look at Conventional Attacks on Scientific Applications


Workshop: 5th International Workshop on Cyber Security in High Performance Computing

Authors: Michael Polinski and Peter Dinda (Northwestern University)

Abstract: How vulnerable are scientific applications to conventional attacks mounted by intentionally manipulating their input? Prior work largely leaves this question unanswered. We begin to answer it by carefully adapting fuzzing and security-focused bug triage to characterize vulnerabilities in scientific codebases. Our process discovers almost 1000 unique crashes (likely distinct bugs) across Enzo, GROMACS, Laghos, and LAMMPS (about 3 million lines of code). Per application, an average of 7.3% of them represent easily exploitable vulnerabilities according to heuristic security analysis. Additionally, 26% of all bugs (51% for GROMACS) are exploitable heap memory safety issues. Cumulatively, we find over 110 distinct vulnerabilities that may permit arbitrary code execution (ACE). Manual inspection confirms the existence of exploitable bugs. Using a coding agent, we generate 59 reproducible proof-of-concept ACE exploits. Our results suggest untrusted data could have catastrophic effects on artifact integrity, creating real risk as computational scientific results drive increasingly consequential societal decisions.


Back to 5th International Workshop on Cyber Security in High Performance Computing Archive Listing

Back to Full Workshop Archive Listing