There is extensive debate about the potential impact of the climate mechanism REDD+ on the welfare of forest-dwelling people. To provide emission reductions, REDD+ must slow the rate of deforestation and forest degradation: such a change will tend to result in local opportunity cost to farmers at the forest frontier. Social safeguard processes to mitigate negative impacts of REDD+ are being developed and can learn from existing safeguard procedures such as those implemented by the World Bank. Madagascar has a number of REDD+ pilot projects with World Bank support including the Corridor Ankeniheny-Zahamena (CAZ). Nearly two thousand households around the corridor have been identified as ‘project affected persons’ (PAPs) and given compensation. We compare households identified as project affected persons with those not identified. We found households with more socio-political power locally, those with greater food security, and those that are more accessible were more likely to be identified as eligible for compensation while many people likely to be negatively impacted by the REDD+ project did not receive compensation. We identify three issues which make it difficult for a social safeguard assessment to effectively target the households for compensation: (a) poor information on location of communities and challenging access means that information does not reach remote households; (b) reluctance of people dependant on shifting agriculture to reveal this due to government sanctions; and © reliance by safeguard assessors on non-representative local institutions. We suggest that in cases where the majority of households are likely to bear costs and identification of affected households is challenging, the optimal, and principled, strategy may be blanket compensation offered to all the households in affected communities; avoiding the dead weight costs of ineffective safeguard assessments. The Paris Agreement in December 2015 recognised REDD+ as a key policy instrument for climate change mitigation and explicitly recognised the need to respect human rights in all climate actions. However, safeguards will be prone to failure unless those entitled to compensation are aware of their rights and enabled to seek redress where safeguards fail. This research shows that existing safeguard commitments are not always being fulfilled and those implementing social safeguards in REDD+ should not continue with business as usual.