Poisoning AI Models: New Frontiers in Data Manipulation Attacks
Kumrashan Indranil Iyer · International Journal of Innovative Research in Computer and Communication Engineering · 2023
Artificial intelligence (AI) and data science models play a crucial role in critical sectors such as cybersecurity, healthcare, and finance, driving key insights and decision-making processes. However, as AI adoption grows, so does its exposure to emerging threats, particularly model poisoning attacks. In these attacks, adversaries stealthily manipulate training data to corrupt model behavior, either causing it to produce malicious outputs or rendering it ineffective against specific threats. This paper examines the methods and motivations behind data poisoning attacks, focusing on how adversaries compromise data pipelines, manipulate model performance, and evade detection. We also examine potential countermeasures and discuss ongoing research challenges that must be overcome to protect AI systems from these evolving threats.