Background Acute myeloid leukemia (AML) is a hematopoietic malignancy which is biologically, phenotypically and genetically very heterogeneous. Outcome of patients with AML remains dismal, highlighting the need for improved, less toxic therapies. Chimeric antigen receptor T-cell (CART) immunotherapies for patients with refractory or relapse (R/R) AML are challenging because of the absence of a universal pan-AML target antigen and the shared expression of target antigens with normal hematopoietic stem/progenitor cells (HSPCs), which may lead to life-threating on-target/off-tumor cytotoxicity. CD33-redirected and CD123-redirected CARTs for AML are in advanced preclinical and clinical development, and they exhibit robust antileukemic activity. However, preclinical and clinical controversy exists on whether such CARTs are myeloablative. Methods We set out to comparatively characterize in vitro and in vivo the efficacy and safety of 41BB-based and CD28-based CARCD123. We analyzed 97 diagnostic and relapse AML primary samples to investigate whether CD123 is a suitable immunotherapeutic target, and we used several xenograft models and in vitro assays to assess the myeloablative potential of our second-generation CD123 CARTs. Results Here, we show that CD123 represents a bona fide target for AML and show that both 41BB-based and CD28-based CD123 CARTs are very efficient in eliminating both AML cell lines and primary cells in vitro and in vivo. However, both 41BB-based and CD28-based CD123 CARTs ablate normal human hematopoiesis and prevent the establishment of de novo hematopoietic reconstitution by targeting both immature and myeloid HSPCs. Conclusions This study calls for caution when clinically implementing CD123 CARTs, encouraging its preferential use as a bridge to allo-HSCT in patients with R/R AML.