A diagonal-unloading (DU)-guided spatial binary masking method is proposed for speech extraction with a moving interferer. The method operates frame by frame from a single-snapshot covariance matrix, avoiding multi-frame covariance estimation, which can be unreliable under time-varying interference. Although DU provides strong directional interference suppression, direct single-snapshot DU can distort target speech. To alleviate this effect, a spatial binary mask is built from partial-DU directional responses steered toward the target and interferer directions of arrival (DOAs) and applied to a reference microphone signal. Simulations with a stationary target and moving interferer show robust performance across signal-to-interference ratios (SIRs), signal-to-noise ratios (SNRs), reverberation times ($T_{60}$), and DOA errors, although attenuation of spatially white noise remains limited.