socialprotection.org (15.01.2025) Artificial Intelligence (AI) has undergone rapid advancements in recent times in terms of computational power and general usage of algorithms across multiple areas, such as healthcare diagnosis, fraud detection and public service delivery. As we will see in this post, usage of AI, particularly algorithms, is not novel in public policy or in social protection and tends to increase (Benoit 2024). The wider acceptance of AI and general utilization of algorithms in public policy serves as the main point of contention. On one hand, as datasets are more widely available, improvements in generative AI take place, and stakeholders are further incentivized into using these tools (Ohlenburg 2020, Benoit 2024). To this end, efficiency is a major objective in using AI, notably on improving the usage of public resources (Murray 2024). On the other hand, efficiency usually comes at a cost, with risks at multiple fronts, be it on transparency or ethics (Bennoit 2024, Vredenburgh 2024), inclusion challenges (Considine et al 2022), among others. Based on this background, we briefly present how AI is (can potentially be) used in social protection, with a focus on automated decision-making. Next, we discuss the risks of exclusion of the most vulnerable, accompanying mitigation measures and good practices.