<p>Relapse in myeloid malignancies remains one of the most persistent barriers to durable therapeutic success, largely due to the survival of leukemic stem cells (LSCs) that function as reservoirs of minimal residual disease. These cells evade eradication through a combination of quiescence, metabolic plasticity, and protection within specialized bone marrow niches, enabling disease re-emergence despite aggressive frontline interventions. Conventional modalities, including cytotoxic chemotherapy, hematopoietic stem cell transplantation, and emerging CAR-T cell therapies, predominantly target proliferating leukemic bulk and often spare relapse-initiating populations. To address this limitation, this study formally defines Root-Cause Immunotherapy as a therapeutic framework that redirects treatment objectives toward the elimination of LSCs rather than transient cytoreduction. Within this framework, root antigens are stratified into three mechanistic classes: mutation-derived neoantigens, stemness-associated markers, and antigens mediating niche interaction and immune evasion. Targeting these nodes enables more selective immunologic engagement of relapse-competent cells. Advances in AI-guided TCR engineering including graph-based modeling, attention-driven architectures, and sequence-level generative design facilitate rational prediction of TCR–peptide–MHC interactions while minimizing off-target reactivity. These approaches are further reinforced by controllability modules such as CRISPR-based receptor editing, regulated degradation circuits, and biosensor-linked safety switches. Concurrent modulation of the bone marrow microenvironment enhances immune accessibility to LSCs, transforming immune-privileged niches into therapeutically permissive domains. This framework is presented as a hypothesis-driven and forward-looking model rather than an established clinical pipeline, with individual components currently supported by preclinical or early-phase clinical evidence.</p>