Human induced pluripotent stem cells (hiPSC) have the potential to generate healthy cells and tissues for the study and medical treatment of a large number of diseases. The utility of putative hiPSC-based therapies is constrained by a lack of robust quality-control assays that address the stability of the cells or their capacity to form teratomas after differentiation. Here we report that virally derived hiPSC, but not human embryonic stem cells (hESC) or hiPSC derived using episomal non-integrating vectors, exhibit a propensity to revert to a pluripotent phenotype following differentiation. This instability was revealed using our published method to identify pluripotent cells undergoing very early-stage differentiation in standard hESC cultures, by fluorescence activated cell sorting (FACS) based on expression of the cell surface markers TG30 (CD9) and GCTM-2. Differentiated cells cultured post-FACS fractionation from virally-derived hiPSC lines re-acquired immunoreactivity to TG30 (CD9) and GCTM-2, formed stem cell-like colonies and re-expressed canonical pluripotency markers. Furthermore, differentiated cells from pluripotency-reverting hiPSC lines generated teratomas in immunocompromised mice, raising concerns about their safety in downstream applications. In contrast, differentiated cell populations from hESC and episomally derived hiPSC did not show any of these abnormalities. Our assays may be used to identify unsafe hiPSC cell lines and this information should be considered when selecting hiPSC lines for clinical use and indicate that experiments using these unsafe hiPSC lines should be interpreted carefully.