AI² Framework: Artificial Intelligence Adversary Imitation Framework
Year: 2025
Type: article
Abstract: Key updates in v1.3: Specified criteria for high-risk AI: established guidance for when to treat systems as high-risk AI, including those capable of becoming high risk, i.e., LLMs with dangerous plugins. Layered AI Decomposition: New guidance on classifying AI systems by Modality, Field, Capability, Model, and Operational Techniques to expose layer-specific risks. AI Solution Layer-Risk Alignment: AI-specific risks mapped to AI layers. Refined Risk Profiling: Enhanced process for identifying, categorizing, and prioritizing over 100 AI-specific risks. Integration with Offensive Security: Expanded tooling and checklists for executing adversarial imitation assessments against high-risk AI systems. Governance & Communication Focus: Added emphasis on stakeholder documentation, cross-functional coordination, and continuous risk monitoring.
Author: Brandon Lee
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6967694071
FWCI:
Cites:
Cited by:
Related to: