Prior work shows that children selectively learn from credible speakers. Yet little is known how they treat information from non‐credible speakers. This research examined to what extent and under what conditions children may or may not learn from problematic sources. In three studies, we found that children displayed trust toward previously inaccurate speakers. Children were equally likely to extend labels from previously accurate and inaccurate speakers to novel objects. Moreover, they expected third parties to share labels provided by previously inaccurate speakers. Only when there was clear evidence that the speakers' information was wrong (as in the case when speakers' perceptual access to the information was blocked), did young children reject the label. Together, the findings provide evidence that young children do not completely ignore the labels supplied by non‐credible speakers unless there is strong reason to do so.