Antimicrobial peptides (AMPs) are promising candidates to address the global antimicrobial resistance crisis, yet their traditional design remains labor-intensive and inefficient. Here, we developed BroadAMP-GPT, an integrated computational-experimental framework that combines AI-driven generation, multi-tiered screening, and experimental validation to rapidly discover potent AMPs with broad-spectrum activity. Using this platform, 57% of AI-generated candidates exhibited potent efficacy against ESKAPE pathogens - key culprits of multidrug-resistant infections. An outstanding candidate, AMP_S13, demonstrated exceptional stability under diverse physiological conditions, including extreme pH (2-10), proteolytic exposure, and elevated temperatures, while maintaining minimal cytotoxicity and low hemolytic activity. AMP_S13 also showed robust efficacy, reducing mortality in infection model and accelerating wound healing in a murine MRSA skin infection model. These results validate BroadAMP-GPT as a transformative tool for accelerating the discovery of stable, broad-spectrum and low-toxicity AMPs, offering a scalable strategy to address the urgent threat of multidrug-resistant pathogens.