Title
Adversarial AI in the cyber domain
Author
Brink, N.W.T.
Kamphuis, Y.N.
Maas, Y.
Jansen-Ferdinandus, G.R.
van Stijn, J.J.
Poppink, B.
de Haan, P.
Chiscop, I.
Publication year
2023
Abstract
What threats are associated with the use of AI? This is a question that TNO seeks to answer through its recent research into the vulnerabilities of AI applications in the cyber domain. Artificial Intelligence (AI) systems use large amounts of data to make decisions in a complex system (AI HLEG, 2020). In order for an AI system to learn specialized tasks, such as discrimination of the different elements within the complex system it operates in (also known as classification), Machine Learning (ML) is applied. These are computer programmes that learn automatically and efficiently through experience (Mitchell, 1997). Besides civilian applications, AI also has a lot of potential in the security domain, as a significant proportion of activities in that area depend on making decisions based on the right information (Swillens, 2022). AI systems are therefore a relevant option to consider for Defence sector applications. For instance, AI currently already plays an important role in information gathering, as well as in driving autonomous and semi-autonomous vehicles such as drones (Xue, Yuan, Wu, Zhang & Liu, 2020; Araya & King, 2022). However, proper analysis of the ability of these AI systems to withstand external threats is essential before they can be deployed on a large scale. TNO is contributing by researching the state of the art when it comes to AI system robustness. This article summarises the conclusions of that research
To reference this document use:
http://resolver.tudelft.nl/uuid:7d48190f-16c9-4a6d-b914-cb8193cf540b
TNO identifier
984412
Report number
TNO 2023 R10292-EN
Publisher
TNO, Den Haag
Document type
report