Danish researchers built an AI system called life2vec that analyzed 6 million people’s complete life histories and predicted four-year mortality with 78% accuracy, outperforming traditional insurance models by 11%. Life insurance companies are developing AI models that ingest “non-traditional data” beyond typical medical exams, including detailed treatment records and lifestyle indicators, with wearable devices and health apps feeding insurers data for claim disputes. Life insurance attorneys warn that carriers might use AI predictions during contestability periods to claim algorithms detected early warning signs applicants “should have disclosed,” leading to application denials or policy cancellations based on patterns invisible to applicants. Sune Lehman, who led the life2vec research, explicitly warns against insurance applications, arguing they would “threaten the basic principle of shared risk,” as AI models might learn that certain zip codes, job types, or income levels correlate with earlier death and embed those biases into coverage decisions. Unlike traditional actuarial tables, AI systems operate as black boxes, making their reasoning nearly impossible to challenge or audit.