An Approach to Measure the Effectiveness of the MITRE ATLAS Framework in Safeguarding Machine Learning Systems Against Data Poisoning Attack

The growth of machine learning systems in critical domains and across society continues with increasing momentum. The future potential influence these systems may have on interactions between machines and humans appears considerable. In this context ensuring the robustness of ML systems against cyber-attacks is crucial. Data poisoning attacks have emerged as a significant threat to the confidentiality, integrity, and availability triad of ML systems. This research presents a comprehensive evaluation of the effectiveness of the MITRE ATLAS framework in protecting ML systems against data poisoning attacks. This project introduces the AERS framework, a comprehensive system designed to assess the effectiveness of the ATLAS framework in protecting ML systems against data poisoning attacks. The AERS framework comprises six distinct categories, each with specific subcategories and qualitative ratings. ATLAS tactics are mapped to the stages of the cyber kill chain, allowing for scenario-based assessments. By aligning threats with real-world attack phases, the AERS framework offers a structured approach to evaluate the robustness of ML defences.