NIST Artificial Intelligence Technology Evaluation Program
Issued by
National Institute of Standards and Technology, Information Technology Laboratory
- September 30, 2026 · Correction — Rewrote the entry as what AITE is: a voluntary NIST testbed launched in July 2026, with no conformance, documentation, or procurement requirements (fact-check finding). (Cody Maxwell)
- October 1, 2026 · Correction — Rewrote the practical steps, newsletter hook, search description, and audience fields to match the entry's corrected content. (Cody Maxwell)
NIST's AI Technology Evaluation (AITE) program, launched in July 2026, is a voluntary testing program. Data providers submit non-public datasets and tasks, model providers submit models, and NIST scores them in a sequestered testbed. It sets no requirements for organizations.
Applies To
Overview
NIST's Technology Test and Evaluation Division, part of the Information Technology Laboratory, launched the AI Technology Evaluation (AITE) program in July 2026. Its overview and road map, version 1.0, is dated 24 July 2026, and the first evaluation period began in August 2026. AITE is a volunteer program that tests AI models on blind data in a sequestered testbed. Data providers submit original datasets and tasks that others cannot see. Model providers submit models to be tested on them, and NIST supplies common data, metrics, and scoring. NIST says the program acts as a neutral third party for hosting tests. Using data that is not public avoids contamination between training and test data. The goal is to measure the state of the art. AITE is one of the AI evaluations run by NIST's Information Technology Laboratory. It is not an extension of the AI Risk Management Framework, and it sets no conformance criteria, documentation duties, or procurement requirements.
Key Requirements
- •Participation is voluntary, through data provider or model provider tracks.
- •Data providers submit original, non-public datasets and tasks.
- •Model providers submit models for testing on those datasets.
- •NIST supplies common data, metrics, and scoring in a sequestered testbed.
- •The program sets no conformance criteria, documentation duties, or procurement requirements.
What Your Organization Must Do
- →Treat AITE as voluntary, and do not record it as a compliance obligation in your control inventory.
- →Decide with engineering whether to join as a data provider, a model provider, or neither.
- →Confirm that submitted datasets are original and non-public before any data provider submission.
- →Check contract and confidentiality terms before sending datasets or models to a third party testbed.
- →Avoid citing AITE scores as evidence of regulatory compliance, since results measure performance only.
- →Keep AITE work separate from NIST AI Risk Management Framework activities, which serve a different purpose.
Playbook Guidance
Step-by-step implementation guidance for compliance teams.
Frequently Asked Questions
- Is NIST AITE legally binding for federal agencies procuring AI systems?
- No, AITE is a volunteer testing program, and NIST's materials include no procurement or conformance requirements.
- How does NIST AITE differ from the NIST AI RMF?
- They are separate: the AI RMF is a voluntary risk management framework, while AITE is a testbed that measures how AI models perform on blind data.
- What does NIST AITE ask of participants?
- Data providers submit original datasets and tasks that stay hidden from others. Model providers submit models, and NIST runs the tests and scores them.
- Can NIST AITE results be used to satisfy AI evaluation requirements in sector-specific regulations?
- NIST's materials do not tie AITE to any regulation. Its results measure model performance and the state of the art, not compliance.
- What penalties apply if an organization's AI system performs poorly in NIST AITE?
- None, because taking part is voluntary. The program has no penalties or consequences for poor results or for not taking part.
- When did NIST AITE start?
- AITE launched in July 2026, and its first evaluation period began in August 2026. There are no conformance deadlines, because it is a voluntary testing program.