Researchers at the University of Pennsylvania presented an AI-enabled, human-in-the-loop framework to identify next-generation CAR T targets with multi-cancer potential. Published in Cell, the approach combines large language models with single-cell RNA-seq datasets to generate and iteratively refine antigen nominations for experimental testing. The team filtered more than 10,000 candidate antigens using CAR T design constraints, ran repeated LLM simulations to reduce noise, and then experimentally validated a top candidate: GPNMB. In preclinical models, GPNMB-directed CAR T cells showed tumor elimination not only in melanoma but also in monoblastic leukemia and colorectal adenocarcinoma. The work focuses on speeding target selection without fully replacing expert review, addressing an enduring bottleneck in solid-tumor CAR T development.