Modern day technology has found its way into every aspect of our lives-- be it the server storing our social information, the hand-held smartphones, the home security systems or a remotely monitored pacemaker. Unfortunately, this also increases the opportunity for agents with malicious intent to violate the privacy, availability or integrity of these applications. In fact, with the advancement of Artificial Intelligence (AI) and faster hardware, the process of finding and exploiting vulnerabilities is no longer as time-consuming as before. Moving Target Defense (MTD) is emerging as an effective technique in addressing these security concerns. This technique, as used by the cyber security community, however, does not incorporate the dynamics of a multi-agent system between an attacker and defender, resulting in sub-optimal behavior. My study of such systems in a multi-agent context helps to enhance the security of MTD systems and proposes a list of challenges for the AI community. Furthermore, borrowing the example of MTD systems from the cyber security community, we can address some security concerns of the present day AI algorithms. In this abstract, I describe my research work that uses AI for enhancing security of a multi-agent MTD system and highlight research avenues in using MTD for enhancing security of present AI algorithms.
[1]
Ananthram Swami,et al.
The Limitations of Deep Learning in Adversarial Settings
,
2015,
2016 IEEE European Symposium on Security and Privacy (EuroS&P).
[2]
Milind Tambe,et al.
From Physical Security to Cyber Security ?
,
2015
.
[3]
Sailik Sengupta,et al.
Moving Target Defense for Web Applications using Bayesian Stackelberg Games: (Extended Abstract)
,
2016,
AAMAS.
[4]
Milind Tambe,et al.
From physical security to cybersecurity
,
2015,
J. Cybersecur..
[5]
Sailik Sengupta,et al.
A Game Theoretic Approach to Strategy Generation for Moving Target Defense in Web Applications
,
2017,
AAMAS.
[6]
Marthony Taguinod,et al.
Toward a Moving Target Defense for Web Applications
,
2015,
2015 IEEE International Conference on Information Reuse and Integration.
[7]
Joan Bruna,et al.
Intriguing properties of neural networks
,
2013,
ICLR.
[8]
John Schulman,et al.
Concrete Problems in AI Safety
,
2016,
ArXiv.
[9]
Fabio Roli,et al.
Adversarial Pattern Classification Using Multiple Classifiers and Randomisation
,
2008,
SSPR/SPR.