You are here

Hackable Intelligence - white paper

Document library
03 April 2020

Why security shouldn’t be overlooked when implementing Artificial Intelligence solutions

 

Many researches have and are proving that AI applications can be easily fooled and hacked so they make wrong decisions and fail in their tasks in ways that benefit potential attackers. Different attacks have been proposed against AI that compromise the confidentiality, integrity and availability of the systems deploying these solutions.

Those attacks against AI are fundamentally different from traditional cyberattacks. This is because the underlying algorithms used to implement AI systems are inherently vulnerable. They can’t be easily patched or replaced as compared to fixing bugs and securing codes when it comes to facing traditional cyberattacks. Furthermore, when compliance programs have been utilized in different industries to protect against traditional cyberattacks, there are no clear standardized guidelines that help industries to implement AI solutions that protect from possible attacks on their AI systems. 
 

Why this publication?

Inspired by these facts, we focus in this article on some of the attacks proposed against AI applications. We also provide insight into best practices that businesses implementing AI solutions can adopt to secure against bad actors. 

 

About the author 

Samraa Alzubi is a Cyber Security Consultant at Approach. Samraa has a master’s in cyber security from ULB university, her master thesis, done last year, researched the attacks against machine learning and proposed a new black-box adversarial reprogramming attack against image classifiers. 

 

Share this publication
focus



 
 



Keep up with our latest news

By clicking on the request button above, you agree our privacy statement