The journalism industry is facing a decisive moment. As generative AI and large language models (LLMs) become entrenched in every facet of information production, newsrooms stand at a crossroads: either build systems that serve the public interest—or remain dependent on commercial platforms that do not. This keynote outlines the case for a journalism-specific LLM: a model trained on licensed, representative news data, designed to support—not replace—journalistic work. Drawing on a blueprint developed in collaboration with technologists, journalists, and researchers, I will explain how such a model could be built, funded, governed, and maintained. I’ll also highlight what’s at stake if we don’t act—loss of editorial independence, diminished transparency, and erosion of public trust. By asserting control over the data and architecture of AI tools, journalism can reclaim its technological agency and deliver better outcomes for newsrooms and the public.