The role of host immunity in emergence of evasive SARS-CoV-2 Spike mutations under therapeutic monoclonal antibody (mAb) pressure remains to be explored. Here, we show that patients treated with various anti-SARS-CoV-2 mAb regimens develop evasive Spike mutations with remarkable speed and high specificity to the targeted mAb-binding sites. Mutations develop more frequently in immunocompromised patients and strongly correlate not only with neutralizing capacity of the therapeutic mAbs, but also with an anti-inflammatory and healing-promoting host milieu. We further built and deploy machine-learning models on host-derived biomarkers that identify patients at high risk of developing escape mutations against therapeutic mAbs with high accuracy. While our data suggest that host-driven responses are essential for development of mutant SARS-CoV-2, the mechanisms and models described here could also be utilized to reduce risk of treatment failure in high-risk populations receiving anti-SARS-CoV-2 mAb treatments and improve mitigation strategies for possible dissemination of escape SARS-CoV-2 mutants.