In UK data, I document the prevalence of misbeliefs regarding the State Pension eligibility age (SPA) and these misbeliefs' predictivity of retirement. Exploiting policy variation, I estimate a lifecycle model of retirement in which rationally inattentive households learning about uncertain pension policy endogenously generates misbeliefs. Endogenous misbeliefs explain 43\%-88\% of the excessive (given financial incentives) drop in employment at SPA. To achieve this, I develop a solution method for dynamic rational inattention models with history-dependent beliefs. Costly attention makes the SPA up to 15\% less effective at increasing old-age employment. Information letters improve welfare and increase employment.