The use of AI for security has proven to be a promising avenue for many critical tasks such as intrusion detection and malware classification.

Reducing manual effort, increasing response times, and generalising to new examples allows defenders to react quickly to incoming threats and allocate valuable human expertise more efficiently.

However, the dynamic nature of data typically seen in security tasks, resulting from the ongoing arms race between attackers and defenders, causes issues that need to be addressed in order to use machine learning to its full potential. Concept drift occurs as malicious techniques evolve and previously trained models aren’t equipped to handle them. The crafting of adversarial examples enables attackers to create specific inputs that bypass ML-based detectors. And many open questions regarding the security of AI, such as information leakage, covert backdoors, and data poisoning, continue to be researched.

AI is a powerful if not revolutionary technology for society, but in reaping the rewards we must also guarantee its safety.

Related Publications

Intriguing Properties of Adversarial ML Attacks in the Problem Space
Fabio Pierazzi*, Feargus Pendlebury*, Jacopo Cortellazzi, Lorenzo Cavallaro
IEEE S&P · 41st IEEE Symposium on Security and Privacy, 2020
@inproceedings{pierazzi2020problemspace,
author = {F. Pierazzi and F. Pendlebury and J. Cortellazzi and L. Cavallaro},
booktitle = {2020 IEEE Symposium on Security and Privacy (SP)},
title = {Intriguing Properties of Adversarial ML Attacks in the Problem Space},
year = {2020},
volume = {},
issn = {2375-1207},
pages = {1308-1325},
doi = {10.1109/SP40000.2020.00073},
url = {https://doi.ieeecomputersociety.org/10.1109/SP40000.2020.00073},
publisher = {IEEE Computer Society},
}
TESSERACT: Eliminating Experimental Bias in Malware Classification across Space and Time
Feargus Pendlebury*, Fabio Pierazzi*, Roberto Jordaney, Johannes Kinder, and Lorenzo Cavallaro
USENIX Sec · 28th USENIX Security Symposium, 2019
@inproceedings{pendlebury2019tesseract,
author = {Feargus Pendlebury*, Fabio Pierazzi*, Roberto Jordaney, Johannes Kinder, and Lorenzo Cavallaro},
title = {{TESSERACT: Eliminating Experimental Bias in Malware Classification across Space and Time}},
booktitle = {28th USENIX Security Symposium},
year = {2019},
address = {Santa Clara, CA},
publisher = {USENIX Association},
note = {USENIX Sec}
}
Transcend: Detecting Concept Drift in Malware Classification Models
Roberto Jordaney, Kumar Sharad, Santanu K. Dash, Zhi Wang, Davide Papini, Ilia Nouretdinov, and Lorenzo Cavallaro
USENIX Sec · 26th USENIX Security Symposium, 2017
@inproceedings {jordaney2017,
author = {Roberto Jordaney and Kumar Sharad and Santanu K. Dash and Zhi Wang and Davide Papini and Ilia Nouretdinov and Lorenzo Cavallaro},
title = {{Transcend: Detecting Concept Drift in Malware Classification Models}},
booktitle = {26th USENIX Security Symposium},
year = {2017},
address = {Vancouver, BC},
url = {https://www.usenix.org/conference/usenixsecurity17/technical-sessions/presentation/jordaney},
publisher = {USENIX Association},
note = {USENIX Sec}
}
DroidSieve: Fast and Accurate Classification of Obfuscated Android Malware
Guillermo Suarez-Tangil, Santanu Kumar Dash, Mansour Ahmadi, Johannes Kinder, Giorgio Giacinto, and Lorenzo Cavallaro
ACM CODASPY · 7th ACM Conference on Data and Application Security and Privacy, 2017
@inproceedings{codaspy17,
author = {Guillermo Suarez-Tangil and Santanu Kumar Dash and Mansour Ahmadi and Johannes Kinder and Giorgio Giacinto and Lorenzo Cavallaro},
title = {{DroidSieve: Fast and Accurate Classification of Obfuscated Android Malware}},
booktitle = {{Proceedings of the Seventh ACM Conference on Data and Application Security and Privacy}},
year = {2017},
month = {March},
url = {http://dx.doi.org/10.1145/3029806.3029825},
doi = {10.1145/3029806.3029825},
note = {ACM CODASPY}
}
Misleading Metrics: On Evaluating Machine Learning for Malware with Confidence
Roberto Jordaney, Zhi Wang, Davide Papini, Ilia Nouretdinov, and Lorenzo Cavallaro
TR@RHUL · Technical Report, 2016
@TechReport{RHUL2016,
author = {Roberto Jordaney and Zhi Wang and Davide Papini and Ilia Nouretdinov and Lorenzo Cavallaro},
title = {{Misleading Metrics: On Evaluating Machine Learning for Malware with Confidence}},
institution = {Royal Holloway, University of London},
year = {2016},
number = {2016-1},
note = {TR@RHUL}
}
DroidScribe: Classifying Android Malware Based on Runtime Behavior
Santanu Kumar Dash, Guillermo Suarez-Tangil, Salahuddin Khan, Kimberly Tam, Mansour Ahmadi, Johannes Kinder, and Lorenzo Cavallaro
IEEE S&P-MoST · IEEE Security and Privacy Workshops: Mobile Security Technologies, 2016
@inproceedings{most16-droidscribe,
author = {Santanu Kumar Dash and Guillermo Suarez-Tangil and Salahuddin Khan and Kimberly Tam and Mansour Ahmadi and Johannes Kinder and Lorenzo Cavallaro},
title = {DroidScribe: Classifying Android Malware Based on Runtime Behavior},
booktitle = {IEEE Security and Privacy Workshops: Mobile Security Technologies},
year = 2016,
month = {May},
note = {IEEE S&P-MoST}
}
Prescience: Probabilistic Guidance on the Retraining Conundrum for Malware Detection
Amit Deo, Santanu Kumar Dash, Guillermo Suarez-Tangil, Volodya Vovk, and Lorenzo Cavallaro
ACM CCS-AISec · 9th ACM CCS Workshop on Artificial Intelligence and Security, 2016
@inproceedings{aisec16,
author = {Amit Deo and Santanu Kumar Dash and Guillermo Suarez-Tangil and Volodya Vovk and Lorenzo Cavallaro},
title = {{Prescience: Probabilistic Guidance on the Retraining Conundrum for Malware Detection}},
booktitle = {9th ACM CCS Workshop on Artificial Intelligence and Security},
year = {2016},
note = {ACM CCS-AISec}
}
Conformal Clustering and Its Application to Botnet Traffic
Giovanni Cherubin, Ilia Nouretdinov, Alexander Gammerman, Roberto Jordaney, Zhi Wang, Davide Papini, and Lorenzo Cavallaro
SLDS · 3rd International Symposium of Statistical Learning and Data Science, 2015
@inproceedings{cherubin,
author = {Giovanni Cherubin and Ilia Nouretdinov and Alexander Gammerman and Roberto Jordaney and Zhi Wang and Davide Papini and Lorenzo Cavallaro},
title = {{Conformal Clustering and Its Application to Botnet Traffic}},
booktitle = {Statistical Learning and Data Sciences, 3rd International Symposium},
year = {2015},
note = {SLDS}
}