This paper addresses the issue of asynchronous brain-switch. The detection of a speciﬁc brain pattern from the ongoing EEG activity is achieved by using the Riemannian geometry, which oﬀers an interesting framework for EEG mental task classiﬁcation, and is based on the fact that spatial covariance matrices obtained on short-time EEG segments contain all the desired information. Such a brain-switch is valuable as it is easy to set up and robust to artefacts. The performances are evaluated oﬄine using EEG recordings collected on 6 subjects in our laboratory. The results show a good precision (Positive Predictive Value) of 92% with a sensitivity (True Positive Rate) of 91%.