A widespread hypothesis in brain imaging posits that neurodegenerative disorders constitute premature aging. Despite its prominence, this brain aging hypothesis (BAH) has not been verified against suitable alternatives. In this work, we first test a key assumption of BAH: Age information is necessary for detecting Alzheimer’s Disease (AD). We compared brain representations that were maximally uninformed about chronological age against ones that were maximally informed about age. We found that absence of aging information impairs AD detection, providing causal evidence for BAH. Second, we investigated whether explicit age modeling confers advantages in transfer learning for AD detection. We evaluated pretraining strategies for age, sex, and BMI inference and found that while pretraining improved representation stability and quality, these tasks converged to similar learned representations with no single phenotype providing superior advantage for neurodegeneration detection. Rather, all fine-tuned models for disease detection implicitly developed increased structural similarity to the age-inference model. These findings demonstrate that aging and neurodegeneration are fundamentally linked, yet aging information emerges naturally during learning of brain features without dedicated encoding. This moves current thinking past brain-age gap conceptualizations and suggests new directions for foundation models integrating richer phenotypic information.