REDD+ aims to offset greenhouse gas emissions through "Reduced Emissions from Deforestation and forest Degradation". Some authors suggest that REDD+ can bring additional benefits for biodiversity, namely for the conservation of extinction-prone restricted-range species. Here, we assess this claim, using São Tomé Island (Democratic Republic of São Tomé and Príncipe) as a case study. We quantified the abundance of bird and tree species, and calculated the aboveground carbon stocks across a gradient of land-use intensity. We found a strong spatial congruence between carbon and the presence and abundance of endemic species, supporting the potential of REDD+ to protect these taxa. We then assessed if REDD+ could help protect the forests of São Tomé and Príncipe. To do so, we used OSIRIS simulations to predict country-level deforestation under two different REDD+ designs. These simulations showed that REDD+ could promote the loss of forests in São Tomé and Príncipe through leakage. This happened even when additional payments for biodiversity were included in the simulations, and despite São Tomé and Príncipe having the fourth highest carbon stocks per land area and the second highest biodiversity values according to the OSIRIS database. These results show weaknesses of OSIRIS as a planning tool, and demonstrate that the benefits that REDD+ might bring for biodiversity are strongly dependent on its careful implementation. We recommend that payment for ecosystem services programmes such as REDD+ develop safeguards to ensure that biodiversity co-benefits are met and perverse outcomes are avoided across all tropical countries. In particular, we advise specific safeguards regarding the conservation of extinction-prone groups, such as island restricted-range species.