Results that do not confirm expectations are generally referred to as 'negative' results. While essential for scientific progress, they are too rarely reported in the literature - Brain-Machine Interface (BMI) research is no exception. This led us to organize a workshop on BMI negative results during the 2018 International BCI meeting. The outcomes of this workshop are reported herein. First, we demonstrate why (valid) negative results are useful, and even necessary for BMIs. These results can be used to confirm or disprove current BMI knowledge, or to refine current theories. Second, we provide concrete examples of such useful negative results, including the limits in BMI-control for complete locked-in users and predictors of motor imagery BMI performances. Finally, we suggest levers to promote the diffusion of (valid) BMI negative results, e.g. promoting hypothesis-driven research using valid statistical tools, organizing special issues dedicated to BMI negative results, or convincing institutions and editors that negative results are valuable.