I propose a text mining model for automatically extracting alliances between firms from news articles. I leverage the pre-trained language model RoBERTa (Liu et al., 2019) and a large amount of labeled examples from the SDC alliance database to fine-tune the model. The resulting system is able to detect alliance announcements in documents, extract the participating firms, and flag alliances according to their purpose. I show that the model is highly accurate in the firm name recognition and relation classification tasks. I run inference on a large corpus of news articles and show that the model can be used to significantly extend existing data sources.