AI² Framework: Artificial Intelligence Adversary Imitation Framework

The AI² Framework introduces a structured, technically rigorous methodology for assessing the security and risk posture of high-risk AI systems through adversarial imitation. Designed to integrate with offensive security programs, this framework uniquely maps AI-specific risks to adversarial testing techniques, leveraging the NIST AI Risk Management Framework, MITRE ATLAS, and leading practices from Microsoft, IBM, and others. It provides actionable guidance for red teams to evaluate generative and agentic AI systems, offering pre-assessment checklists, risk-to-tactic mappings, and threat modeling tailored to both model and application layers. The AI² Framework serves as both a practical implementation guide and a comprehensive risk reference, enabling organizations to adopt AI technologies responsibly while addressing emerging threats, complex attack surfaces, and evolving regulatory expectations. This Version 1.0 release delivers a full workflow to support continuous risk identification, targeted adversarial assessment, and informed control implementation.