Suicide is inherently difficult to predict. Epidemiological research identified many general risk factors such as a depression, but these predictors have limited predictive power. Machine learning offers a set of tools that can combine hundreds of predictors resulting in the most optimal prediction. It might therefore offer a powerful way to predict inherently complex behaviour such as suicide. In a recent study, state of the art ML algorithms where applied to a large Swedish dataset of 126.205 patients treated in psychiatry containing over 400 potential risk factors. Although the presented results such as an area under the curve if 88% sounds promising, many questions on for example the cost of a false negative remain unanswered. In our comment, we critically discuss the presented findings, and bring up some unanswered questions.