The LHC is one of our most powerful tools for discovering new physics, but existing search strategies are blind to many possible new physics scenarios. One of the most challenging possibilities are Soft Unclustered Energy Patterns (SUEPs). SUEPs can arise in confining dark sector models with pseudo-conformal running of the dark coupling constant, and manifest as highly isotropic showers of low-energy hadrons in the laboratory frame, making classical collider searches extremely challenging. Unsupervised machine learning has recently gained popularity as a tool to search for new physics that may evade standard searches. In this talk I will first provide an overview of how unsupervised machine learning is used for anomaly detection in collider physics applications. Then, I will present a newly developed unsupervised search strategy that can effectively probe SUEP at the LHC and be applied to a variety of scenarios. I will demonstrate that our methodology provides sensitivity to SUEPs produced in exotic Higgs decays down to O(1%) branching ratio.