Washington, D.C. — TechNet, the national, bipartisan network of technology CEOs and senior executives, submitted comments from President and CEO Linda Moore to the National Institute of Standards and Technology (NIST) on the public draft of its “TEVV-Athlon Framework for Evaluating AI Systems,” proposed standards for testing, evaluation, verification, and validation (TEVV) of artificial intelligence models. In the letter, Moore encouraged NIST to preserve the framework’s flexible, risk-based foundation while providing clearer and more practical guidance on how organizations should design evaluations, interpret and report results, allocate responsibilities, test safeguards, and determine when reevaluation is necessary. Stated Moore:
“TechNet appreciates the progress NIST has made on crafting a practical framework for AI testing and evaluation. This framework is an essential tool for understanding how AI systems perform, identifying potential risks, and ensuring appropriate safeguards are in place before and after deployment. As AI systems become more capable, a TEVV framework can help organizations measure performance, assess whether safeguards are working, and respond to emerging risks. We look forward to continued engagement with NIST and other stakeholders to establish clear, flexible, and risk-based practices that strengthen public trust in AI while enabling continued innovation and adoption.”
TechNet’s comment letter offered detailed recommendations on the following aspects of NIST’s TEVV-Athlon Framework:
- Preserve a risk-based and scalable framework;
- Clarify core TEVV terminology;
- Connect evaluation results to decisions and acceptance criteria;
- Make results reproducible and fit for comparison;
- Protect the integrity of the evaluation process;
- Clarify shared responsibility across the AI value chain;
- Expand agentic and system-level evaluation;
- Evaluate the effectiveness of safeguards;
- Establish change triggers for lifecycle reevaluation;
- Promote interoperability and evidence reuse; and
- Provide additional templates and worked examples.