Cancer stem cells undergo epithelial-mesenchymal transition (EMT) to drive metastatic dissemination in experimental cancer models. However, tumour cells undergoing EMT have not been observed disseminating into the tissue surrounding human tumour specimens, leaving the relevance to human cancer uncertain. Here, we identify an EMT stem cell state that retains EpCAM and CD24 after undergoing EMT and exhibits enhanced plasticity. This afforded the opportunity to investigate whether retention of EpCAM and CD24 alongside upregulation of the EMT marker Vimentin can identify disseminating EMT stem cells in human oral cancer specimens. Examining disseminating tumour cells in the stromal region of 3500 imaging fields from 24 human oral cancer specimens, evenly divided into metastatic and non-metastatic specimens, we see a significant enrichment of EpCAM, CD24 and Vimentin co-stained cells in metastatic specimens. Through training an artificial neural network on the EpCAM, CD24 and Vimentin co-staining, we predict metastasis with high accuracy (F1 0.91; AUC 0.87). We have observed, for the first time, disseminating EMT stem cells in patient histological specimens and demonstrated their utility for predicting metastatic disease.