Vasileios KemerlisAug 12, 2023

Brown Secure Systems Lab (https://gitlab.com/brown-ssl/) had a strong representation @usenixassociation sponsored events this year! Neophytos Christou presented IvySyn at USENIX Security Symposium (SEC) '23, while Di Jin talked about EPF at USENIX Annual Technical Conference (ATC) '23 earlier in July!

IvySyn fuzzes Deep Learning (DL) frameworks (TensorFlow, PyTorch) for memory-safety bugs and automatically synthesizes Python code snippets for triggering the respective vulnerabilities | https://www.usenix.org/.../usenixse.../presentation/christou | https://gitlab.com/brown-ssl/ivysyn

EPF (ab)uses the (e)BPF interpreter for bypassing various kernel hardening mechanisms in Linux -- we also introduce a set of lightweight defenses against EPF-style attacks | https://www.usenix.org/conference/atc23/presentation/jin | https://gitlab.com/brown-ssl/epf

#brownssl #ivysyn #epf #usenix #atc23 #usesec23

Brown Secure Systems Lab · GitLab

https://brown-ssl.slack.com

GitLab
Vasileios KemerlisMar 20, 2023
📢 Our work on automated discovery of memory safety vulnerabilities in Deep Learning (DL) frameworks has been accepted at USENIX Security
2023! Joint work with Neophytos Christou, Di Jin, Vaggelis Atlidakis, and Baishakhi Ray (Columbia) | https://arxiv.org/abs/2209.14921 | https://gitlab.com/brown-ssl/ivysyn | 39 CVEs 😎 🤘 💣 | #ivysyn #brownssl #usenixsecurity #usesec23
IvySyn: Automated Vulnerability Discovery in Deep Learning Frameworks

We present IvySyn, the first fully-automated framework for discovering memory error vulnerabilities in Deep Learning (DL) frameworks. IvySyn leverages the statically-typed nature of native APIs in order to automatically perform type-aware mutation-based fuzzing on low-level kernel code. Given a set of offending inputs that trigger memory safety (and runtime) errors in low-level, native DL (C/C++) code, IvySyn automatically synthesizes code snippets in high-level languages (e.g., in Python), which propagate error-triggering input via high(er)-level APIs. Such code snippets essentially act as "Proof of Vulnerability", as they demonstrate the existence of bugs in native code that an attacker can target through various high-level APIs. Our evaluation shows that IvySyn significantly outperforms past approaches, both in terms of efficiency and effectiveness, in finding vulnerabilities in popular DL frameworks. Specifically, we used IvySyn to test TensorFlow and PyTorch. Although still an early prototype, IvySyn has already helped the TensorFlow and PyTorch framework developers to identify and fix 61 previously-unknown security vulnerabilities, and assign 39 unique CVEs.

arXiv.org