Throughout the 21st century, the increasing presence of Artificial Intelligence (AI) in everyday applications underscores its crucial role in enhancing community comfort and convenience. And with the world’s increased use in technology, there is no doubt that Artificial Intelligence will remain as a persistent center of focus on the stage of technological advancement. However, it is crucial to understand how the inner workings of AI components could be manipulated for malicious use by threat actors. Large Language Models (LLMs) are a type of Artificial Intelligence solution that process data, can recognize patterns and generate output text. Through large and extensive training, LLMs have the ability to produce natural language text as a response – but at what cost? The use of LLMs is prone to security risks and data breaches, as manipulated data that is fed to these systems could ultimately lead to incorrect, false, or unintended outputs. Overall, the vulnerabilities in the input data can compromise the integrity of information produced by the model and introduce unforeseen privacy attacks. This research paper summarizes the latest findings on the security risks that are associated with LLMs. In addition, the paper explores the inner workings of LLMs, the advantages and limitations of the use in this model, and recommendations to address the risks.
Kawle et al. (Sat,) studied this question.