Adversarial machine learning attacks (AMLAs) is an umbrella term for a variety of cyber-attacks that revolve around machine learning systems, most often impairing the latter’s functioning or obtaining new information in an unauthorised manner. One of the key characteristics of machine learning systems is that they adapt themselves to the queries, results and feedback they receive while operating; and this continuous drive towards improvement is what makes them so valuable, but also vulnerable.
The sixth CC-DRIVER policy brief presents the nature of AMLAs, human drivers and legal challenges, and discusses current EU law and AI regulation.