"Microsoft and MITRE Create Tool to Help Security Teams Prepare for Attacks on Machine Learning Systems"
Microsoft and MITRE have developed a plug-in that integrates multiple open-source software tools to better prepare cybersecurity professionals for attacks on Machine Learning (ML) systems. The Arsenal tool implements tactics and techniques defined by the MITRE ATLAS framework. It has been built off of Microsoft's Counterfit library as an automated adversarial attack library, so security professionals can accurately emulate attacks on ML-containing systems without needing a deep ML or Artificial Intelligence (AI) background. Counterfit is a tool developed by Microsoft that enables ML researchers to conduct a range of adversarial attacks against AI algorithms. MITRE CALDERA is a tool for creating and automating adversary profiles. MITRE ATLAS is a knowledge platform with adversary tactics, methodologies, and case studies for ML systems based on real-world observations, demonstrations from ML red teams, and more. Using Microsoft's Counterfit library, the Arsenal plug-in allows CALDERA to simulate adversarial attacks and behaviors. The tool currently contains a limited number of adversary profiles based on publicly available information. This article continues to discuss the new Arsenal plug-in that brings together Microsoft Counterfit, MITRE CALDERA, and MITRE ATLAS to help cybersecurity practitioners better understand threats to ML systems.