Purpose – This study maps the intellectual foundations, conceptual structure and research fronts of artificial intelligence (AI) in financial risk research and explains the field's shift from prediction towards financial risk intelligence. Design/methodology/approach – A Web of Science search returned 3,399 records; document-level screening produced a core corpus of 2,161 articles and reviews. Performance analysis, keyword co-occurrence, thematic mapping, reference publication year spectroscopy, co-citation, co-authorship, bibliographic coupling and a validated dictionary-based content analysis (Cohen's κ = 0.79) were combined, with robustness checks. Findings – Output grew by 32.3% a year over 2018–2025. The knowledge base combines financial-distress models and machine-learning methodology. Four stable streams emerge: explainable credit-risk machine learning; AI, FinTech, market and systemic risk; intelligent classification for bankruptcy and credit scoring; and corporate distress. Explainability rose from 5% to 22% of abstracts, whereas out-of-time validation (6%), calibration (3%) and fairness (2%) remain rare. Research limitations/implications – The corpus is Web of Science-based and finance-facing, so computer-science venues are under-represented. Practical implications – Banks, insurers and supervisors should jointly validate interpretable baselines, nonlinear learners, text and network signals, calibration and fairness. Originality/value – The study separates the knowledge base, research front and emerging frontier, and defines financial risk intelligence as a decision-level construct distinct from AI risk management, model risk governance and explainable AI, with five testable propositions.