How Hard is Trojan Detection in DNNs? Fooling Detectors With Evasive Trojans

Mantas Mazeika, Andy Zou, Akul Arora, Pavel Pleskov, Dawn Song, Dan Hendrycks, Bo Li, David Forsyth

Primary Area: societal considerations including fairness, safety, privacy

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: trojan detection, neural trojans, trojans, hidden functionality, monitoring, security, ML safety

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: We investigate making trojan attacks in DNNs more evasive for model-level detectors.

Abstract:

Trojan attacks can pose serious risks by injecting deep neural networks with hidden, adversarial functionality. Recent methods for detecting whether a model is trojaned appear highly successful. However, a concerning and relatively unexplored possibility is that trojaned networks could be made harder to detect. To better understand the scope of this risk, we develop a general method for making trojans more evasive based on several novel techniques and observations. In experiments, we find that our evasive trojans reduce the efficacy of a wide range of detectors across numerous evaluation settings while maintaining high attack success rates. Surprisingly, we also find that our evasive trojans are substantially harder to reverse-engineer despite not being explicitly designed with this attribute in mind. These findings underscore the importance of developing more robust monitoring mechanisms for hidden functionality and clarifying the offense-defense balance of trojan detection.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 7765