At the Netherlands Forensic Institute (NFI), we've developed a Python-based deep learning model to spot life-threatening messages in lawfully intercepted communication data, like those from the infamous chat service Encrochat. Thanks to the application of our model in collaboration with the Dutch Police, dozens of potential victims of violent crimes, including murder, serious assault, and kidnapping, have been warned and safeguarded. In this talk, we'll dive into the development, implementation, and success of our deep learning model in the fight against violent criminal activities. We'll also tackle the risks tied to using deep learning for these cases and discuss the precautions we took to ensure responsible and accurate use.