Human cancers arise through the sequential acquisition of somatic mutations that create successive clonal populations. Human cancer evolution models could help illuminate this process and inform therapeutic intervention at an early disease stage, but their creation has faced significant challenges. Here, we combined induced pluripotent stem cell (iPSC) and CRISPR-Cas9 technologies to develop a model of the clonal evolution of acute myeloid leukemia (AML). Through the stepwise introduction of three driver mutations, we generated iPSC lines that, upon hematopoietic differentiation, capture distinct premalignant stages, including clonal hematopoiesis (CH) and myelodysplastic syndrome (MDS), culminating in a transplantable leukemia, and recapitulate transcriptional and chromatin accessibility signatures of primary human MDS and AML. By mapping dynamic changes in transcriptomes and chromatin landscapes, we characterize transcriptional programs driving specific transitions between disease stages. We identify cell-autonomous dysregulation of inflammatory signaling as an early and persistent event in leukemogenesis and a promising early therapeutic target.