What is TEVV for AI?
TEVV is test, evaluation, verification and validation. NIST's AI RMF uses it for the checks run on an AI system at every stage of its lifecycle.
TEVV stands for test, evaluation, verification and validation. In AI, it is the term the US National Institute of Standards and Technology (NIST) uses for the checks that show whether an AI system works as intended and what risks it carries. NIST’s AI Risk Management Framework (AI RMF) says these tasks “are performed throughout the AI lifecycle” (Appendix A), not once before release.
Where TEVV appears in the AI RMF
The AI RMF has four functions: Govern, Map, Measure and Manage (AI RMF Core). The core text and Playbook tie TEVV to Map, Measure and Manage.
| Function | What it says about TEVV | Source |
|---|---|---|
| Map | Map 2.3: “Scientific integrity and TEVV considerations are identified and documented, including … construct validation.” | AI RMF Core |
| Measure | Measure 2.1: “Test sets, metrics, and details about the tools used during TEVV are documented.” Measure 2.13: “Effectiveness of the employed TEVV metrics and processes in the measure function are evaluated and documented.” | AI RMF Core |
| Manage | A suggested action under Manage 1.1: “Utilize TEVV outputs from map and measure functions when considering risk treatment.” One under Manage 4.1: “Perform post-deployment TEVV tasks to evaluate AI system validity and reliability”, along with bias, privacy and security. | Playbook, Manage |
The Measure function also says AI systems “should be tested before their deployment and regularly while in operation” (AI RMF Core). Test results feed the decisions made in Manage, so TEVV is the evidence behind a risk decision.
What TEVV covers at each lifecycle stage
Appendix A of the AI RMF describes TEVV tasks by lifecycle stage (source).
| Stage | TEVV tasks named in the AI RMF |
|---|---|
| Design, planning and data | Internal and external validation of assumptions for system design, data collection and measurements relative to the intended context of deployment |
| Development | Model validation and assessment |
| Deployment | System validation and integration in production, with testing and recalibration; user experience; compliance with existing legal, regulatory and ethical specifications |
| Operations | Ongoing monitoring for periodic updates, testing and recalibration of models by subject matter experts; tracking reported incidents and errors; detecting emergent properties and their impacts; redress and response processes |
Tests can be planned as early as design. Appendix A also says that, ideally, “AI actors carrying out verification and validation tasks are distinct from those who perform test and evaluation actions.” NIST’s Measure text adds that “processes for independent review can improve the effectiveness of testing” (AI RMF Core).
Test, evaluation, verification and validation
The AI RMF names the four activities and groups them in pairs, but gives no formal definition of each word. The pairing is the useful part. Test and evaluation produce the data: test sets, metrics and results. Verification and validation are the separate judgments made from that data. The AI RMF asks teams to keep the two sets of people apart where they can.
For the practical meaning of testing an LLM application, see LLM evaluation. For runtime checks that act on a model’s inputs and outputs rather than measuring it, see AI guardrails.
NIST’s TEVV-Athlon draft
NIST AI 200-2, “The TEVV-Athlon Framework for Evaluating AI Systems,” is an initial public draft. NIST describes it as “a four-stage method for developing customized assessments of AI systems based on organizational TEVV objectives.” In the draft’s model, AI systems “are tested via a set of Events and Tools” that produce data on Blocks tied to the measurement concepts an organization cares about. It is meant to cover statistical machine learning models, large language models, multimodal models and agentic systems.
NIST announced the draft on 2026-08-07 with a 60-day comment period that closed on 2026-10-06 (NIST). The draft is not final, so check the NIST page and the draft PDF for the current text.
A note on other uses of the term
Search results for “TEVV” also include US defense material on test and evaluation of autonomous systems, such as a Defense Innovation Marketplace page titled “Test & Evaluation, Validation & Verification”. It says the goal of its work is to “enable more rapid and cost-effective fielding of autonomous technology.” The NIST usage above is the one that applies to AI risk management for software systems such as LLM applications.
Producing TEVV evidence with HoneyHive
The AI RMF asks that test sets, metrics and tool details be documented (Measure 2.1). In HoneyHive, an evaluation runs a dataset through your application and records the results of your evaluators, which gives you a stored record of the test set, metrics and outputs for each run. Whether that record satisfies a particular requirement is for you and your assessors to decide.
FAQ
What does TEVV stand for?
Test, evaluation, verification and validation. The NIST AI Risk Management Framework spells it out in its Measure function, which expects "objective, repeatable, or scalable test, evaluation, verification, and validation (TEVV) processes" to be in place, followed and documented (AI RMF Core).
Is TEVV the same as T&E?
The two terms overlap, and the AI RMF does not define the difference. NIST's AI RMF treats test and evaluation as one pair of actions and verification and validation as another, and says the people doing the second pair should ideally be different from those doing the first (Appendix A).
What is the TEVV-Athlon framework?
A NIST draft, NIST AI 200-2, that describes a four-stage method for building customized assessments of AI systems from an organization's own TEVV objectives. NIST announced the initial public draft on 2026-08-07 and its comment period closed on 2026-10-06 (NIST).