The selection of genetically engineered immune or hematopoietic cells in vivo after gene editing remains a clinical problem and requires a method to spare on-target toxicity to normal cells. Here, we develop a base editing approach exploiting a naturally occurring CD33 single nucleotide polymorphism leading to removal of full-length CD33 surface expression on edited cells. CD33 editing in human and nonhuman primate hematopoietic stem and progenitor cells protects myeloid progeny from CD33-targeted therapeutics without affecting normal hematopoiesis in vivo, thus demonstrating potential for improved immunotherapies with reduced off-leukemia toxicity. For broader application to gene therapies, we demonstrate highly efficient (>70%) multiplexed adenine base editing of the CD33 and gamma globin genes, resulting in long-term persistence of dual gene-edited cells with HbF reactivation in nonhuman primates. Using the CD33 antibody-drug conjugate Gemtuzumab Ozogamicin, we show resistance of engrafted, multiplex edited human cells in vivo, and a 2-fold enrichment for edited cells in vitro. Together, our results highlight the potential of adenine base editors for improved immune and gene therapies.