The impact of the life cycle of large language models on cybersecurity

In recent years, there has been an increase in the use of neural networks for generating synthetic content, which goes hand in hand with the rise of new cybersecurity challenges. Generative models can have various impacts on cybersecurity, ranging from positive to negative.A significant area is the security of deployment and operation of generative models, primarily large language models.

The goal of this thesis is to identify problem areas in a selected field of LLM deployment and operation (e.g., model inference attacks, model theft, or information theft) and analyze new trends, approaches, defenses, and their characteristics, impacts, and potential applications. The work should then propose new protection methods based on the analysis and research on the state of security for the selected areas.

Recommended areas of work focus:
Attacks and defenses in the side channel domain of large language models
Attacks and defenses in the inference domain of large language models
Defences in the area of adversarial attacks on large language models

Participation in relevant international conferences and publication in peer-reviewed or scientific journals are expected.