In a world dominated by AI, even the most sophisticated technology can be easily misled or manipulated by adversaries. The National Institute of Standards and Technology (NIST) and collaborators have identified vulnerabilities in AI and machine learning (ML) and outlined various adversarial tactics that can be used to confuse or “poison” these systems, with no fail-safe defense available.
The publication, titled “Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations,” is part of NIST’s efforts to support the development of trustworthy AI and can provide valuable insights for AI developers and users on potential attacks and strategies to mitigate them, despite the absence of robust defenses.
The report classifies attacks into four major types: evasion, poisoning, privacy, and abuse attacks, each with specific goals and characteristics, and provides mitigating approaches for each, emphasizing that the existing defenses against adversarial attacks are incomplete, and there remain significant theoretical problems in securing AI algorithms.