Deep learning is a common tool used to detect cyberattacks. Models trained with this technology can identify complex patterns which is ideal for distinguishing between benign and malicious behaviors. However, a drawback of using these models is that they require a massive amount of data to train them. Moreover, to perform classification, the data must be labeled by an expert which can be prohibitively expensive or simply impractical to accomplish at scale.  In this talk we will take a look at how one kind of architecture, the autoencoder, has revolutionized the way we can detect cyber threats in computer networks. First, we will cover the basics of how these models work and how they are used to perform anomaly detection. Then we will explore how they can be used as practical solutions for detecting malicious network traffic. Finally, we will discuss how we can use them to identify man-in-the-middle and supply-chain attacks through echo analysis.