Method for enhancing the robustness of large language models against adversarial attacks

Abstract

This paper presents a method for the authentication of information resources based on block cipher encryption, designed to ensure the authenticity, integrity, and confidentiality of data during storage, processing, and transmission in computer systems and networks. The relevance of improving cryptographic authentication mechanisms is substantiated in the context of the growing number of cyber threats associated with data forgery, unauthorized access, and the compromise of information resources. An analysis of contemporary authentication methods is conducted, revealing that existing solutions do not always provide an appropriate balance between cryptographic strength, processing speed, and computational efficiency. The application of block cipher algorithms for generating authentication data is examined, and the key requirements for constructing the proposed method are identified. A method for authenticating information resources is proposed, based on cryptographic key generation, block transformation of the information resource, authentication code generation, integrity verification, and decision-making regarding the authenticity of the resource. Unlike existing approaches, the proposed method integrates encryption and authentication procedures, enables localization of modified data blocks, provides quantitative authentication assessment, and supports risk-oriented decision-making. The sequence of cryptographic processing of the information resource is formalized, and models for constructing an integrated authentication profile, assessing authenticity and integrity, and selecting an appropriate response according to the risk of compromise are developed. The conducted structural and functional evaluation demonstrated the capability of the proposed method to localize modified data blocks and quantitatively determine the degree of deviation of the information resource from its reference state. The obtained results confirm the feasibility of further software implementation and experimental validation of the proposed method in modern information and communication systems.