As Generative Adversarial Networks (GANs)can be trained to mimic any distribution of data, their potential for both good and evil is downright immense, if not boundless. This is particularly important in the cybersecurity sphere where it raises the question whether we can indeed truly trust data. To answer that question and raise awareness of the potential and risks that GANs bring to our digital world, let's have a quick look at how those neural networks are already put to use in today's cyber defense arena.
The emergence of a new digital technology is often followed by new cybersecurity threats. To be able to recognise and prevent áll malicious attacks, it is therefore necessary that we also learn to address the other side of the coin: the risks for data, businesses and individuals that stem from those new digital technologies themselves.
Deep neural networks to the rescue
One such technology is the GAN, short for Generative Adversarial Network. A deep neural network that operates primarily on image data, it can just as easily be used to create fully new datasets of any kind (video, audio, text, images) that can furthermore be implemented in any industry.
Due to this key feature and benefit, the GAN machine learning technique, developed in 2014, is already widely in use today. It is used, for instance, to improve the ability to prevent cyber attacks from infecting a computer system or device. As the examples below will show, this cyber defense also involves taking active steps to anticipate adversarial cyber actions and to counter intrusions.
Financial fraud detection and prevention
The use of GAN algorithms can contribute to the fight against cybercrime, preventing financial fraud and money laundering schemes. Currently, the approach to identify suspicious financial transactions is based on human-engineered rules and a large set of databases. But GANs can advance that approach if properly and specifically trained to identify suspicious financial activities that could otherwise remain under the radar. This strategy has already been used to great effect by Swedbank, a Swedish bank that trained the algorithms on its specific dataset. This resulted in the trained model being able to predict if new financial transactions were fraudulent or not.
Malware detection and threat prevention
Another way for GANs to boost your cyber defense strategy is by contributing to malware detection and preventing malicious attacks. An example in case is Defense-GAN, a new defense mechanism against adversarial attacks. Trained to model the distribution of original, unperturbed images, this new security method can be used with any classification model and against any type of attack.
Additionally, GANs can strengthen your cyber defense strategy by helping to improve detection systems for different types of malware. GANs research has shown, for instance, that by augmenting the training set with generated adversarial examples, the classifier is able to detect more malware families than by using other approaches.
Data protection implementation
Lastly, GANs help to apply the CIA principles of confidentiality, integrity, and availability onto data. It is therefore considered a privacy-enhancing method. By using GANs, synthetic data is created that is considered as anonymous. It can be used for different purposes, such as testing or, in the medical domain, for anonymising medical records. In the latter example, it protects the sensitive data of patients and reduces the risks of negative effects, should that data get leaked or stolen. Certainly, using artificially created data rather than real data is a more secure option to avoid other malicious uses against real persons, such as impersonations.
Eager to learn more about the opportunities and risks of GANs from a privacy and data protection perspective? Read what our Data Protection Consultant has to say on the subject.