An AI model called TRisk demonstrated superior accuracy in predicting blood clots and bleeding risks for atrial fibrillation patients, achieving 82% discrimination accuracy compared to 71% for the standard CHA2DS2-VASc score across 411,850 UK patients and 16,218 US patients. The model analyzes temporal patient data to predict 12-month thromboembolic and bleeding events, outperforming current clinical tools like HAS-BLED and ORBIT for bleeding prediction. This precision medicine approach represents a significant advancement in cardiovascular risk stratification, potentially transforming how clinicians decide which atrial fibrillation patients need blood thinners. The AI could reduce unnecessary anticoagulant prescriptions by 7-8% while maintaining stroke prevention effectiveness, generating substantial healthcare savings. However, this remains a preprint study awaiting peer review, so results may change. The model's ability to capture evolving patient risk over time addresses a key limitation in current guidelines. While promising for personalized medicine, the real-world implementation challenges and long-term clinical outcomes require further validation through randomized controlled trials before widespread adoption.