Incident investigations in mining are frequently completed under time pressure, with evidence dispersed across witness statements, photographs, operational systems, monitoring data, maintenance records and controlled documents. Investigation quality therefore depends not only on the analytical method used, but on how consistently evidence is collected, challenged and linked to findings. Variation in practice can weaken organisational learning and make it difficult to demonstrate why conclusions and corrective actions are supported by evidence. This extended abstract describes an artificial intelligence (AI, OpenAI, 2026) assisted investigation workflow designed to structure evidence and support investigator decision-making without transferring accountability to the AI system. The approach uses a controlled large language model (LLM) workflow to organise evidence, identify gaps and discrepancies, develop candidate timelines, prompt consideration of performance shaping factors, and support evidence-linked contributing factors and corrective actions. Human confirmation remains mandatory for material findings and final report content. This is consistent with systems-based investigation principles that seek latent organisational conditions rather than defaulting to individual error (Reason, 1997), and with AI risk-management concepts emphasising governance and oversight (Tabassi, 2023; Autio et al, 2024).