It is inevitable that Hackers will collectively embrace and abuse AI systems. After all, that is the mindset that differentiates hackers: the desire and capacity to take a piece of technology and twist it to fit our needs. There has been a lot of talk about generative models in the context of social engineering and operational security, but comparatively little discussion of other ways that AI systems might be used as part of a hack. This talk will discuss a recent engagement in which we had to attack a system that interfaced with, among other things, a security camera. We will explain how we implemented a technique that we call a Model in the Middle (ModITM) attack, in which an ML model was inserted as a malicious interloper between a pair of services in order to degrade system integrity. Additionally, we will discuss how using an ML model in an attack changes the design parameters of both the model and the attack infrastructure. Then, we will generalize the ModITM attack and explain how it can be used in different attack scenarios. To the best of our knowledge, this attack is the only publicly available example of using a machine learning model as a component of an attack chain.