Pharmaceutical quality relies on the proactive detection of process, product, and compliance risks, yet critical signals are often hidden within unstructured sources such as regulatory guidance, deviation narratives, CAPA records, inspection observations, and other quality documents, which are typically reviewed in a fragmented manner. Traditional quality risk management depends heavily on manual document review and local expertise, making it challenging to identify recurring issues across sites, benchmark internal deviations against external regulatory expectations, or develop a comprehensive view of emerging risks. This article proposes an AI-powered regulatory text mining system that ingests regulatory guidelines, deviation reports, and CAPA records to extract risk entities and their relationships, link them to manufacturing processes, and build a queryable quality-risk knowledge graph. The framework integrates document ingestion, preprocessing, named entity recognition, relation extraction, transformer-based risk classification, knowledge graph construction, and dashboard-based decision support, with human verification to ensure interpretability, auditability, and compliance with regulatory standards. By converting scattered textual information into actionable quality-risk intelligence, the system enables quality teams to anticipate compliance gaps, prioritize CAPA activities, and respond more rapidly to evolving regulatory expectations, shifting pharmaceutical organizations from reactive documentation toward predictive, science-based quality oversight.