ABOUT ATTACKS ON LARGE FUNDAMENTAL MODELS
Грибунин Вадим Геннадьевич, Sergey A. Mayorov, Alexander A. Murashko · Voprosy kiberbezopasnosti · 2025
Purpose of the study: to study the possibilities and limitations of an information security attacker in organizing an attack on large fundamental models designed to work with program code. Methods of research: comparison and juxtaposition, system analysis. Results: the article presents the features of large fundamental models as objects of information protection, the principles of implementing the most relevant attacks on large fundamental models designed to work with program code, provides metrics that allow comparing the effectiveness of various approaches to attacks, and identifies the problems that exist in this area for attackers. Scientific novelty: large fundamental models in the world are just beginning to be used for working with program code. The article systematically describes information security threats and possible attacks on systems using these models. Problematic issues requiring further research are also presented.