Overview
Severity: MEDIUM | Affected: U.S. AI Safety Institute | Category: tool
The U.S. AI Safety Institute (USAISI), under the Department of Commerce, has released its first major deliverable: the 'GARLAND' framework. GARLAND (Generative AI Red-teaming and Layered Analysis for National Defense) is a comprehensive, open-source framework designed to standardize the security evaluation of AI models used in critical infrastructure and government applications. The framework includes a suite of automated tools for probing models for vulnerabilities like prompt injection, data poisoning, and privacy leakage, alongside a structured methodology for manual red-teaming exercises. To encourage adoption, USAISI is launching a certification program for security professionals who demonstrate proficiency with the framework. This move is seen as a significant step towards creating a common language and set of standards for assessing AI safety and security before models are deployed in sensitive environments.