This paper examines the potential negative externalities of foreign aid projects: that is, costs that accrue to people outside the aid transaction between the recipient state (the implementer) and the aid financier (the agent). Related literature mostly focuses on governance relationships between agents and powerful donor countries (principals), but they do not explain compliance with policies to prevent negative aid externalities. Instead, due to genuine capacity challenges, incomplete contracts between the agent and implementer, and the latter's political incentives, compliance with agents' social and environmental risk management policies mostly relates to recipient state capacity. To test the hypothesis, I compile a new dataset on states' project-level compliance with World Bank safeguard policies on involuntary resettlement, indigenous peoples, and environmental protection. Statistical support for the hypothesis suggests availability and representativeness biases in how the media and previous literature approach safeguard compliance; a need for the literature to focus more on incomplete agent-implementer contracts; and a more limited role for principal-agent relations in foreign aid.