Content-based information flow control realized by large language model based on neural networks

Abstract An information system is composed of two types of entities, subjects like users and objects like databases from the security viewpoint. A subject issues an operation to an object to manipulate data in the object. An object is protected from malicious accesses of unauthorized subjects in the AC (Access Control) models. Nevertheless, if data in an object $$o_j$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub> <mml:mi>o</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:math> is stored in another object $$o_k$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub> <mml:mi>o</mml:mi> <mml:mi>k</mml:mi> </mml:msub> </mml:math> , a subject $$s_i$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub> <mml:mi>s</mml:mi> <mml:mi>i</mml:mi> </mml:msub> </mml:math> which is allowed to read $$o_k$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub> <mml:mi>o</mml:mi> <mml:mi>k</mml:mi> </mml:msub> </mml:math> can read the data of $$o_j$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub> <mml:mi>o</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:math> in $$o_k$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub> <mml:mi>o</mml:mi> <mml:mi>k</mml:mi> </mml:msub> </mml:math> even if $$s_i$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub> <mml:mi>s</mml:mi> <mml:mi>i</mml:mi> </mml:msub> </mml:math> is not granted an access right of $$o_j$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub> <mml:mi>o</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:math> , i.e. illegal information flow from $$o_j$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub> <mml:mi>o</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:math> to $$s_i$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub> <mml:mi>s</mml:mi> <mml:mi>i</mml:mi> </mml:msub> </mml:math> occurs. In our previous studies, the O-IFC (Object-based Information Flow Control) is proposed where operations occurring illegal information flow are prohibited. Here, a unit of data exchanged among entities is an object. Even if some operations do not occur illegal information flow, the operations are prohibited in the O-IFC. In order to reduce the operations unnecessarily prohibited, a novel C-IFC (Content-based IFC) is proposed where a finer unit of data than an object is considered to be exchanged among entities. In this article, an object is composed of natural language sentences. It is critical to decide whether or not a sentence $$st_j$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mi>s</mml:mi> <mml:msub> <mml:mi>t</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:mrow> </mml:math> in $$o_j$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub> <mml:mi>o</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:math> is the same as another sentence $$st_k$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mi>s</mml:mi> <mml:msub> <mml:mi>t</mml:mi> <mml:mi>k</mml:mi> </mml:msub> </mml:mrow> </mml:math> in $$o_k$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub> <mml:mi>o</mml:mi> <mml:mi>k</mml:mi> </mml:msub> </mml:math> . In this article, the sentence classifier is used which is generated by using the large language model based on the neural networks. If $$st_j$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mi>s</mml:mi> <mml:msub> <mml:mi>t</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:mrow> </mml:math> and <jats:altern