Purpose <p>To synthesise current applications of Artificial Intelligence (AI) in orthopaedic imaging, summarise diagnostic performance, and outline priorities for safe implementation within an Orthopaedic AI Literacy (OAIL) framework.</p> Methods <p>This systematic review was conducted in accordance with PRISMA 2020 guidance. PubMed, Scopus, and Embase were searched for studies published from January 2010 to January 2025, with the most recent searches run on 25 September 2025 and limited to publications up to January 2025. Eligible studies included diagnostic accuracy, feasibility, and implementation research evaluating AI tools on orthopaedic imaging modalities, including radiographs, computed tomography, magnetic resonance imaging, and ultrasound. Data extracted included study design, setting, imaging modality, clinical task, AI approach (deep learning, machine learning, or general AI), reference standard, and performance metrics (for example: sensitivity, specificity, area under the receiver operating characteristic curve (AUC), accuracy, and F1-score). Studies were grouped into key clinical domains. We performed a descriptive domain-level synthesis summarising reported performance metrics as descriptive ranges and summary statistics; pooled estimates were extracted only when reported in published meta-analyses (no de novo pooling), due to heterogeneity and incomplete 2 × 2 data.</p> Results <p>A total of 664 studies were included. Fracture detection and trauma imaging represented the largest subdomain, followed by spine and degenerative disorders, musculoskeletal oncology, arthroplasty and implant assessment, and workflow applications. Deep learning dominated most clinical tasks. Across domains, AUC values commonly ranged from 0.90 to 0.95 and reported sensitivities and specificities were frequently between 80% and 95% (descriptive ranges; not pooled). Most studies were retrospective and single-centre, with limited external validation and under-reporting of calibration, clinical impact, and generalisability, particularly in low- and middle-income settings.</p> Conclusion <p>AI is most mature for fracture detection and degenerative spine imaging, with emerging but less validated applications in implant assessment, oncology, and operational workflow. Evidence limitations constrain generalisability, and AI should be positioned as decision support rather than a replacement for expert judgement. The Orthopaedic AI Literacy (OAIL) framework is proposed to guide clinician training and the safe implementation of AI in orthopaedics. The OAIL framework is normative and educational rather than outcome-predictive, and requires future empirical validation.</p>