New SMART taxonomy classifies AI model access risks for security

A new taxonomy called SMART (Signal-based Model Access Risk Taxonomy) has been introduced to classify attacker access to AI systems based on the information signals they can obtain. This framework aims to provide a more deployment-oriented perspective than traditional white-box, gray-box, and black-box classifications. By organizing evasion attacks according to the richness of available signals, SMART helps organizations make more informed decisions regarding AI procurement and deployment security. AI

IMPACT Provides a structured framework for understanding and mitigating AI system vulnerabilities based on signal exposure.

RANK_REASON The cluster contains a research paper detailing a new taxonomy for AI security. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Hugging Face
• Signal-based Model Access Risk Analysis for AI System Operations Security
• SMART

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →