Authors:
Tiberiu Boros
1
;
Andrei Cotaie
1
;
Antrei Stan
1
;
Kumar Vikramjeet
2
;
Vivek Malik
2
and
Joseph Davidson
2
Affiliations:
1
Adobe Systems, Romania
;
2
Adobe Systems, U.S.A.
Keyword(s):
Machine Learning, Living-off-the-Land (LotL), Feature Engineering, Artificial Intelligence, Random Forest, Commands, CommandLine, OpenSource, Linux.
Abstract:
Among the methods used by attackers to avoid detection, living off the land is particularly hard to detect. One of the main reasons is the thin line between what is actually operational/admin activity and what is malicious activity. Also, as shown by other research, this type of attack detection is underrepresented in Anti-Virus (AV) software, mainly because of the high risk of false positives. Our research focuses on detecting this type of attack through the use of machine learning. We greatly reduce the number of false detection by corpora design and specialized feature engineering which brings in-domain human expert knowledge. Our code is open-source and we provide pre-trained models.