We analyze a natural experiment generated by the interaction of the Social Security DI and OA programs at Full Retirement Age, when DI beneficiaries are automatically converted from the DI program to the OA retired worker program.  At conversion benefit payments continue unchanged, however the DI program’s high implicit marginal tax rate on earnings is abruptly relaxed.  We use administrative Social Security data for the universe of primary worker DI beneficiaries from the 1934-1942 birth cohorts observed in panel over the period of 1995-2008. Our estimates imply that the DI program depresses labor supply among even the oldest DI beneficiaries. In the context of the literature to date that has sought to establish an upper bound on the earnings losses caused by the presence of the DI program by using quasi-experimental variation occurring at the program entry margin, our use of quasi-experimental variation arising from the program exit margin, when individuals are already in their mid-60s and the dominant trend in labor force participation in the population at large is downward, suggests that our estimates are most appropriately viewed as a lower bound estimate of the residual work capacity of all beneficiaries.