An AI model was developed to predict chemotherapy-induced cardiotoxicity (CTRCD) from baseline electrocardiograms (ECGs). The model, called AI-CTRCD, was trained using transfer learning from an existing AI model (AI-EF) that detects reduced left ventricular ejection fraction (LVEF) from ECGs. The AI-CTRCD model was tested on 1011 patients treated with anthracyclines, with 8.7% experiencing CTRCD. High AI-CTRCD scores were associated with a higher risk of CTRCD (hazard ratio 2.66; 95% CI 1.73–4.10). This risk remained consistent after adjusting for known risk factors. The AI-CTRCD score improved prediction of CTRCD beyond traditional factors, with a time-dependent AUC of 0.78 for 2 years compared to 0.74 without the score. The AI-CTRCD model was validated across various subgroups, including different cancer types, sex, baseline LVEF, and anthracycline doses, showing consistent performance. The model also demonstrated robustness in predicting CTRCD even when baseline LVEF was normal. The AI-CTRCD score significantly improved the prediction of CTRCD compared to models without the score, with higher specificity and positive predictive value. The model was also effective in detecting CTRCD under limited echocardiogram capacity, allowing for more efficient use of resources. The study highlights the potential of AI to predict CTRCD from baseline ECGs, which could improve early detection and management of cardiotoxicity in cancer patients. The AI-CTRCD model provides a reliable and accessible tool for risk stratification, potentially reducing the number of patients with CTRCD who are missed during surveillance. The model's performance was consistent across different institutions and was not significantly affected by competing risks such as death. The study underscores the importance of integrating AI-based risk assessment into clinical practice to optimize the management of patients undergoing cardiotoxic chemotherapy.An AI model was developed to predict chemotherapy-induced cardiotoxicity (CTRCD) from baseline electrocardiograms (ECGs). The model, called AI-CTRCD, was trained using transfer learning from an existing AI model (AI-EF) that detects reduced left ventricular ejection fraction (LVEF) from ECGs. The AI-CTRCD model was tested on 1011 patients treated with anthracyclines, with 8.7% experiencing CTRCD. High AI-CTRCD scores were associated with a higher risk of CTRCD (hazard ratio 2.66; 95% CI 1.73–4.10). This risk remained consistent after adjusting for known risk factors. The AI-CTRCD score improved prediction of CTRCD beyond traditional factors, with a time-dependent AUC of 0.78 for 2 years compared to 0.74 without the score. The AI-CTRCD model was validated across various subgroups, including different cancer types, sex, baseline LVEF, and anthracycline doses, showing consistent performance. The model also demonstrated robustness in predicting CTRCD even when baseline LVEF was normal. The AI-CTRCD score significantly improved the prediction of CTRCD compared to models without the score, with higher specificity and positive predictive value. The model was also effective in detecting CTRCD under limited echocardiogram capacity, allowing for more efficient use of resources. The study highlights the potential of AI to predict CTRCD from baseline ECGs, which could improve early detection and management of cardiotoxicity in cancer patients. The AI-CTRCD model provides a reliable and accessible tool for risk stratification, potentially reducing the number of patients with CTRCD who are missed during surveillance. The model's performance was consistent across different institutions and was not significantly affected by competing risks such as death. The study underscores the importance of integrating AI-based risk assessment into clinical practice to optimize the management of patients undergoing cardiotoxic chemotherapy.