SECURITY IN ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING SYSTEMS ANALYSIS OF ADVERSARIAL ATTACKS AND DEFENSE STRATEGIES

Abstract

Artificial Intelligence (AI) and Machine Learning (ML) have played fundamental roles in digital transformation, being widely used in sectors such as healthcare, finance, and transportation. However, these technologies present significant vulnerabilities, especially in relation to adversarial attacks, which manipulate input data to compromise the effectiveness and security of models. In addition, the use of biased data to train algorithms can perpetuate prejudices and inequalities, raising ethical and social concerns. This work aims to analyze the impacts of adversarial attacks on AI and ML systems, as well as propose defense strategies to ensure the reliability and security of these technologies. At the same time, we seek to explore the ethical challenges related to the use of biased data, suggesting guidelines that promote greater equity and responsibility. The research adopts a bibliographic methodology, based on the analysis of scientific articles, technical reports, and case studies. This approach allows identifying vulnerabilities, exemplifying adversarial attacks, and evaluating existing solutions to mitigate their effects. This work is based on authors who explore the use of artificial intelligence in several areas, such as natural language (VASWANI et al., 2017; DEVLIN et al., 2019), speech recognition (RAVANELLI et al., 2020), agriculture (SILVA et al., 2018) and autonomous vehicles (FENG et al., 2021), vulnerability to adversarial attacks (HOSPEDALES et al., 2020). The contributions of Transformers and deep learning (OTTER et al., 2020) stand out. The results indicate that measures such as the use of robust algorithms, the diversification of training data and the implementation of continuous adversarial testing are effective in protecting AI and ML systems. In addition, the need for regulations that ensure transparency and ethics in the development of these technologies is highlighted. It is concluded that, to ensure security and fairness, it is essential to combine technical strategies with ethical guidelines, promoting the responsible adoption of AI and ML.