This study evaluates whether a standard 12‑lead ECG, augmented by a convolutional neural network, can screen for asymptomatic left ventricular dysfunction (ALVD; EF ≤35%). Using paired ECG–echocardiogram data from 97,829 unique Mayo Clinic patients (train 35,970, validation 8,989, test 52,870) selected from 625,326 ECG–TTE pairs acquired within 14 days, the model was trained and tuned on the development sets and assessed on a held‑out test set. Performance metrics included AUC, sensitivity, specificity, accuracy, and prognostic analyses using follow‑up echocardiography. On the independent test cohort, the AI‑ECG achieved AUC 0.93 with sensitivity 86.3%, specificity 85.7%, and accuracy 85.7% for detecting EF ≤35%; with a sensitivity‑targeted threshold, sensitivity was 89.1% and specificity 83.0%, and among patients without comorbidities, AUC rose to 0.98 with accuracy 92.5%. Notably, individuals with a normal contemporaneous EF but a positive AI screen (“false positives”) had a fourfold higher risk of developing EF ≤35% over time (HR 4.1; 95% CI 3.3–5.0), with cumulative incidence ≈9.5% at 5 years versus ≈1.8% in true negatives, suggesting subclinical disease detection, negative predictive value was ~99%. The discussion underscores that an AI‑enabled ECG—ubiquitous and low‑cost—performs on par with or better than several accepted screening tests, outperforms BNP‑based screening (and is less sensitive to age/sex), and could extend beyond human interpretation, while acknowledging limits including modest PPV at the EF ≤35% cutoff, single‑system data, and non-simultaneous ECG–TTE acquisition.