NIST Outlines Next Steps for US Federal AI Standards
The U.S. National Institute of Standards and Technology (NIST) is focusing on creating "coherent federal approaches" to AI risk management and operationalizing trustworthy AI. In a recent update, the agency emphasized the need for active participation from industry, academia, and international partners to develop robust and harmonized AI standards. This signals a shift from establishing principles to creating actionable technical criteria and testable benchmarks.
- This work is a direct result of the October 2023 Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence, which mandated NIST to develop guidelines for AI safety and security. The agency's responsibilities include setting standards for red-team testing, creating guidance for authenticating AI-generated content, and establishing the U.S. Artificial Intelligence Safety Institute (AISI). - A key component of NIST's strategy is the AI Risk Management Framework (AI RMF), first released in January 2023. This voluntary framework is designed to be flexible and non-sector-specific, helping organizations integrate trustworthiness considerations into the entire AI lifecycle. - To address the rapid advancements in generative AI, NIST has launched a dedicated public working group and developed a specific "Generative AI Profile" for the AI RMF. This profile identifies unique risks posed by generative models, such as hallucinations and misuse for disinformation, and proposes specific actions to manage them. - NIST is also developing a plan for global engagement on AI standards, aiming to align its domestic frameworks with international efforts and promote shared standards for AI safety and testing. The roadmap specifically mentions aligning with ISO/IEC standards like ISO/IEC 42001. - The U.S. government's approach, favoring a voluntary, collaborative framework with the private sector, contrasts with the European Union's more comprehensive and legislative AI Act. This reflects a broader geopolitical competition, particularly with China, where the development of international technology standards is a strategic priority. - As part of the implementation, the newly formed U.S. AI Safety Institute (AISI) at NIST will facilitate the development of testing standards and environments for researchers to evaluate emerging AI risks. This includes a new research program called Assessing Risks and Impacts of AI (ARIA) to improve assessment methods. - The legal basis for NIST's work extends from the National Artificial Intelligence Initiative Act of 2020 and is further reinforced by the CHIPS and Science Act of 2022. These acts direct NIST to create a risk-mitigation framework, support the development of trustworthy AI standards, and establish testbeds. - A companion roadmap to the AI RMF outlines future activities, including the development of more use-case-specific profiles, guidance on human-AI teaming, and expanded tools and benchmarks for evaluating AI system trustworthiness.
