In information systems, to prevent illegal information flow is critical. In our previous studies, the sets of objects whose data flow to subjects and objects are considered. Illegal Information flow is defined based on these sets. However, even if a subject issues an operation to an object, data may not flow among objects. Therefore, data in objects have to be checked to know whether or not the data of an object flow to another object. In this paper, we consider the large language model to check information flow among objects. Suppose text data in objects are manipulated by subjects. The fine tuned large language model to make clear the similarity between the text data of \(o_1\) and the text data of \(o_2\) is considered. Based on the fine tuned large language model, the similarity between the text data of \(o_1\) and the text data of \(o_2\) is made clear. If the similarity is large between the text data and \(o_1\) is included in the set of objects whose data flow to \(o_2\) , the text data of \(o_1\) is regarded as to flow to \(o_2\) . On the other hand, if the similarity is small, the text data of \(o_1\) is regarded as not to flow to \(o_2\) even if \(o_1\) is included in the set of objects whose data flow to \(o_2\) . Therefore, the illegal information flow is defined more precisely based on the fine tuned large language model. In this paper, we propose the information flow control based on the fine tuned large language model.