This thesis explores the emerging intersection of large language models (LLMs) and open-source intelligence (OSINT), investigating how generative AI can augment the collection, analysis, and dissemination of publicly available information. While traditional OSINT workflows rely heavily on manual data triage, translation, and cross-referencing, LLMs offer the ability to rapidly summarize large text corpora, extract entities, translate across languages, and present structured narratives from unstructured data. By integrating LLMs into OSINT pipelines, analysts can potentially accelerate sense-making, reduce cognitive load, and focus their efforts on high-value interpretation and verification rather than repetitive processing.
At the same time, the integration of LLMs into intelligence work introduces new challenges around reliability, transparency, and ethics. Models are prone to hallucination, bias, and lack of source attribution, which risks undermining the evidentiary standards required in intelligence practice. Furthermore, reliance on third-party AI services may raise concerns around data security, privacy, and compliance with legal frameworks. This research aims to evaluate both the opportunities and limitations of LLM-assisted OSINT through a combination of literature review, prototype development, and case studies, with the ultimate goal of proposing a framework for responsible adoption of LLMs in intelligence analysis.