This work addresses a two-sided problem: the potential of machine learning to both enhance and degrade IoT security. Machine learning methods can be applied to analyze large amounts of data that IoT devices produce. Such research may reveal unknown patterns in network traffic, detect anomalies that are typical of hacks, and predict emerging threats. These machine learning models provide us with the opportunity to predict and enhance IoT network security before any future hacking occurs. However, the same machine learning technologies can be exploited by cyber-criminals to implement more sophisticated hacks. It is hence essential to be aware of the abuse aspect when employing machine learning towards effective IoT cybersecurity.