As economic uncertainty and external shocks like climate change intensify, China’s Targeted Poverty Alleviation (TPA) policy has become increasingly important in strengthening rural households’ resilience to future poverty risks. Drawing on China Family Panel Studies (CFPS) data from 2012 to 2020, this study applies a Difference-in-Differences (DID) model to identify how TPA reduces farm households’ vulnerability to expected poverty. Moreover, we employ a machine-learning Causal Forest approach to uncover how the policy operates across different levels of household livelihood capital. The analysis shows that TPA lowers expected poverty vulnerability. Mechanism and heterogeneity analysis reveals that TPA reduces the risk of future poverty by strengthening social integration and endogenous development momentum. Notably, households with mid-level land assets, limited housing wealth, and particular demographic structures benefit the most. These findings contribute to contemporary literature streams concerning how livelihood capacity shapes anti-poverty outcomes by providing guidance for developing countries seeking inclusive poverty-reduction strategies.