In 2025, the U.S. National Institute of Standards and Technology (NIST) launched the “AI Standards Zero Draft” initiative to accelerate the formation of international AI standards. Under this initiative, NIST first selects topics with a solid scientific foundation and established research results, then solicits broad input from the community and industry. Based on the collected feedback, NIST drafts a comprehensive “Zero Draft,” which is submitted to Standards Development Organizations (SDOs) as a starting point for discussion, helping avoid delays in reaching consensus when starting from scratch.
The process consists of five steps:
1. Community participation and input: Collect suggestions and needs from various stakeholders.
2. NIST selects and refines the topic: Define the focus and scope.
3. Release of a concept paper or outline: Provide the structure and preliminary content.
4. Release of the complete draft: Consolidate feedback into a full initial draft.
5. Submission to the Standards Development Organization (SDO) : Enter the formal consensus process.
One of the first topics in this initiative focuses on Testing, Evaluation, Verification, and Validation of AI systems—collectively referred to as TEVV. Rather than specifying detailed technical requirements, the draft provides a high-level framework to help different organizations design AI testing and validation approaches based on their needs.
The draft pays particular attention to the complexity and uncertainty of AI. Today’s AI systems and products often comprise multiple layers of technology and data. Algorithms are not easily interpretable, and factors such as training data, system architecture, upstream and downstream relationships, and human-machine interaction contexts all affect performance and risk assessment. Therefore, NIST recommends that TEVV processes consider validity, reliability, sampling methods, resource constraints, and the system’s performance under different environmental conditions.
In addition, the draft emphasizes the importance of governance and process management. Before conducting TEVV, organizations should clearly define objectives and requirements, establish repeatable and traceable testing procedures, and cover aspects such as safety, privacy, misuse prevention, and performance. Some high-level ethical or societal impact issues may not be fully measurable with existing methods and may require interdisciplinary approaches and ongoing monitoring.
After the public consultation period, this draft is expected to be submitted to ISO/IEC JTC 1/SC 42 for discussion. For Taiwan, this represents an important opportunity to participate in global AI standards development and to enhance the international recognition and competitiveness of its AI products. AIEC will continue to monitor the progress of this draft.
