Confabulation is characterized by a propensity to claim false memories as true and is commonly attributed to cognitive monitoring failures. We examined whether such monitoring failures can be mitigated by explicit task instructions aimed to emphasize specific monitoring strategies. Using the Deese-Roediger-McDermott (DRM) paradigm, which robustly elicits false memories in healthy participants, we compared the tendency to endorse unpresented distractor items as studied in three groups of subjects: confabulating patients, non-confabulating patients, and healthy controls. All subjects participated in three conditions: no warning, retrieval warning (warning instructions after study but before test), and encoding warning (warning instructions before study). Control participants reduced false alarms in both warning conditions, but more so under retrieval warning. Although neither patient group showed a reduction in false alarms, confabulators' confidence was altered as a result of warning. These results are discussed in light of the monitoring differences between groups.