The National Institute of Standards and Technology (NIST) is currently collecting feedback on its AI Standards Zero Drafts project. The initiative aims to establish voluntary standards for AI documentation and testing without introducing federal regulations. The deadline for input on the public-facing documentation guidance is this Wednesday, while comments for the TEVV-Athlon framework are accepted until October 6.
The proposed documentation standards cover AI model and dataset disclosures, including version history, training data, and performance metrics. A second draft outlines a framework for testing, evaluation, verification, and validation (TEVV). This four-stage method is designed for various AI systems, ranging from traditional machine learning models to large language models and agentic AI.
For financial institutions, these standards could simplify the procurement process by providing a consistent checklist for vendor evidence. Currently, banks must develop individual diligence processes to compare disparate vendor claims. Standardized documentation would function similarly to a nutrition label, allowing risk teams and auditors to evaluate system capabilities and limitations effectively.
Although the NIST proposals are voluntary, they may influence commercial practices. Financial institutions frequently incorporate technical standards into vendor contracts, potentially making these guidelines a practical requirement for doing business. Additionally, insurers could utilize these standardized records to better assess liability exposure and evaluate controls for AI companies.
