As the landscape of artificial intelligence development and use rapidly evolve, the dual use of AI in both defending and compromising digital and connected infrastructure becomes a key priority. We explore the realm of adversarial attacks to AI systems, a sophisticated and emerging threat vector that challenges the integrity and reliability of AI systems. Adversarial attacks are able to cleverly manipulate AI models to malfunction or produce erroneous outputs, posing significant risks to cybersecurity frameworks that rely on these technologies.
This talk will provide an overview of adversarial AI, illustrating how seemingly robust and functional systems can be subtly and effectively undermined. We will explore notable examples of adversarial attacks on AI, including alteration of input data that causes AI-driven systems to overlook or misclassify threads. The presentation will also discuss the methodologies behind crafting such attacks, their potential impacts, and the strategic measures necessary to detect and mitigate these hidden threats. Through understanding these vulnerabilities, we can better prepare our defenses against the cunning exploits that adversarial AI represents in the cybersecurity domain.
