Peptide antibiotics are an abundant and synthetically tractable source of molecular diversity, but they are often cationic and can be cytotoxic, nephrotoxic and/or ototoxic, which has limited their clinical development. Here we report structure-guided optimization of an amphipathic peptide, arenicin-3, originally isolated from the marine lugworm Arenicola marina. The peptide induces bacterial membrane permeability and ATP release, with serial passaging resulting in a mutation in mlaC, a phospholipid transport gene. Structure-based design led to AA139, an antibiotic with broad-spectrum in vitro activity against multidrug-resistant and extensively drug-resistant bacteria, including ESBL, carbapenem- and colistin-resistant clinical isolates. The antibiotic induces a 3–4 log reduction in bacterial burden in mouse models of peritonitis, pneumonia and urinary tract infection. Cytotoxicity and haemolysis of the progenitor peptide is ameliorated with AA139, and the ‘no observable adverse effect level’ (NOAEL) dose in mice is ~10-fold greater than the dose generally required for efficacy in the infection models.