Abstract: Deep engineering in high-strength brittle rocks is prone to sudden failure, which often triggers severe rock mass hazards. Early warning relies heavily on reliable identification of precursory signals. However, existing acoustic-emission (AE) single-parameter warning methods cannot simultaneously accommodate failure-feature differences among lithologies, limiting warning accuracy and timeliness. In this study, coarse-grained granite, fine-grained granite, and quartzite were investigated using uniaxial compression tests, AE monitoring, and nuclear magnetic resonance (NMR) characterization of pore structures. The lithology-controlled mechanical responses, failure modes, and microfracturing evolution mechanisms were systematically revealed. On this basis, a multi-parameter AE precursor indicator system integrating correlation dimension, b-value, ring count, and average energy centroid frequency (AECF) was established, and the pre-peak evolution patterns and physical implications of each parameter were quantitatively analyzed. The results show that precursory AE signals in highly brittle rocks (fine-grained granite and quartzite) emerge significantly later than those in quasi-brittle rock (coarse-grained granite), and AECF is more sensitive to sudden failure. Accordingly, a lithology-adaptive, stage-wise multi-parameter joint warning strategy is proposed, enabling differentiated identification of failure in rock masses with different brittleness. These findings provide a theoretical basis and a technical pathway for intelligent assessment and real-time early warning of catastrophic risks in deep rock engineering.