ITI Offers Recommendations to NIST on AI Red-Teaming, Stakeholder Engagement, and Standards Development
WASHINGTON – Today, global tech trade association ITI responded to National Institute of Standards and Technology (NIST)’s request for information related to NIST’s assignments in U.S. President Biden’s Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence.
In its comments, ITI recommends that:
- NIST work with international counterparts while developing a generative AI Risk Management Framework (RMF) to increase alignment of approaches across borders, including leveraging international standards;
- Consider the important roles of developers and deployers, as well as others across the AI ecosystem, in advancing transparency for generative AI systems;
- Ensure that evaluation and auditing are scoped to the level of risk an AI system poses;
- Recognize that cybersecurity red-teaming is different from AI red-teaming, differentiate between red-teaming an AI model vs. red-teaming an AI system, and avoid construing red-teaming as a catch-all solution for risk management;
- Consider a flexible and risk-based approach that allows for the use of multiple content authentication techniques appropriate to the level of risk and use case; and
- In developing a global AI standards development and engagement plan, maintain international consistency to the extent possible and continue to recognize the importance of stakeholder engagement in standards development bodies.
“We believe that a generative AI RMF or profile, standards and guidelines for AI red-teaming and model evaluation, and a plan to engage in and advance the development of international standards are all integral to advancing the development and deployment of trustworthy AI. We look forward to continuing to collaborate with NIST as they progress these efforts and encourage NIST to view ITI as a resource,” ITI wrote in its comments.
Read ITI’s full comments here.
ITI has reiterated that safe and responsible AI development and deployment must be grounded in trust, transparency, ethics, and collaboration among government, industry, civil society, and academia. INCITS, an affiliate division of ITI, serves as the U.S. technical advisory group to the international standards body that recently published a standard for managing the risks and opportunities of AI. This management system standard supports improved quality, security, traceability, transparency, and reliability of AI applications.
In January, ITI unveiled a new guide for global policymakers, Authenticating AI-Generated Content: Exploring Risks, Techniques & Policy Recommendations, aiming to address the pressing need to authenticate AI-generated content, including chatbots, image, and audio generators.