← All topics
WORKFLOW STEP 02 / 05

AI Vendor Assessment

Ask for evidence about third-party models, data and controls.

Sources reviewed through

View 6 resources
CURATED COLLECTION06original-source resources

RECOMMENDED RESOURCES

Choose a starting point.

Compare type, format and framework. Every link opens the original publisher.

Official sourceNIST AI Resource Center

AI RMF Playbook: Manage 3

OUR PICK FORDefining supplier control expectations

Suggested actions for monitoring third-party AI resources and documenting risk controls.

Type
Official Guidance
Format
Web
Framework
NIST AI RMF
Access
Free access
Visit original
TemplateBrandQuill

AI Vendor Assessment Workbook

OUR PICK FORComparing vendor responses in a spreadsheet

A 50-question workbook with weighted scoring, evidence links, gaps and a dashboard.

Type
Template
Format
XLSX
Framework
General
Access
Free access
Visit original
Official sourceUK Government / Home Office approach

AI Assurance Questionnaire (v4.3)

OUR PICK FORBuilding an assurance question set

An interactive navigator of 146 questions across governance and AI lifecycle stages; the page describes a redrafted Home Office assurance approach.

Type
Checklist
Format
Web
Framework
General
Access
Free access
Visit original
Official sourceWashington Technology Solutions

Initial Procurement Guidelines for Generative AI

OUR PICK FORQuestions for generative AI suppliers

Includes a sample generative AI procurement questionnaire covering the technology and vendor.

Type
Checklist
Format
PDF
Framework
General
Access
Free access
Visit original
Official sourceUK Government

Portfolio of AI Assurance Techniques

OUR PICK FORUnderstanding possible assurance evidence

Examples of assurance techniques that can inform what evidence to request from AI suppliers.

Type
Official Guidance
Format
Web
Framework
General
Access
Free access
Visit original
ToolVerifyWise / GitHub

VerifyWise

OUR PICK FORA platform-oriented alternative to spreadsheets

Source-available AI governance platform under BSL 1.1, with supplier, system and model management features. Check its license before reuse.

Type
Tool
Format
Web
Framework
NIST AI RMF; ISO 42001; EU AI Act
Access
Free access
Visit original

Quick overview

WHAT IT IS

A structured review of an external AI supplier and the evidence behind its claims.

WHO NEEDS IT

Procurement, security, privacy and AI governance teams.

WHEN TO USE IT

Before shortlisting or approving a vendor, then at material change or renewal.

Before requesting supplier evidence, use AI use case intake to agree on the intended use and data. Use the AI governance framework to identify which risks the supplier review must cover.

Use the AI governance policy template to record who can accept a material evidence gap. Carry the findings into AI procurement so the buying team can turn open gaps into acceptance criteria or contract conditions. Link approval conditions to the AI system inventory for follow-up.

Written by AI Gov Library · Editorial guidance reviewed 7 October 2026. Source review dates describe our checks, not publisher updates. Suggested fields and decisions below are our own starting points; adapt them to your organization.

Compare AI vendor assessment resources

An AI vendor assessment reviews whether a supplier and its product fit a specific use case. It should connect questions to evidence and to a decision. The same supplier may be suitable for drafting internal text but unsuitable for a sensitive decision; assess the proposed use, not just the company.

We suggest choosing a question source for discovery and a separate record for answers, gaps and decisions. Do not treat a numerical score, a framework label or an unanswered questionnaire as approval.

Pick the assessment aid that fits your review
ResourceUse it forProduceWatch for
NIST AI RMF Playbook: ManageFraming ongoing third-party AI risk management.Control expectations and questions about supplier dependencies.Suggested actions, not a vendor certification or scoring tool.
BrandQuill AI Vendor Assessment WorkbookComparing supplier responses in a spreadsheet.An evidence-backed comparison of suppliers and gaps.Review its weighting before use; an aggregate score can hide a critical gap.
UK AI Assurance QuestionnaireSelecting assurance questions across governance and lifecycle stages.A tailored question set for your use case.The navigator is broad; select applicable questions and document why others are out of scope.
Washington Technology Solutions GenAI procurement guidelinesStarting questions for generative AI suppliers.Questions about the specific GenAI service being proposed.Government-oriented guidance; adapt responsibilities to your organization.
UK Portfolio of AI Assurance TechniquesUnderstanding possible ways to substantiate a claim.A request for suitable assurance evidence.An assurance technique must fit the claim and deployment context.
VerifyWiseExploring a platform approach to supplier and system records.A view of whether managed workflows suit your team.Software does not perform the review for you; check its license and operational needs.

The source cards below retain formats and original links. For one supplier, a simple evidence log may suffice. For several suppliers, a workbook can help keep answers comparable. Adopt a platform only when ownership and recurring review work justify it.

Back to page sections ↑

Run an evidence-based supplier review

  1. Fix the assessment scope. Name the product, version, intended task, users, data and deployment arrangement. List exclusions so that a response about one service is not silently applied to another.
  2. Select questions by exposure. Begin with the question set below. Add specialist questions for the data, decisions or integrations involved. Request the same baseline evidence from each shortlisted supplier.
  3. Record answers and evidence separately. A claim is not its supporting document. Record the evidence date, product scope, access restrictions and reviewer. For unavailable evidence, state the gap rather than guessing.
  4. Test the material claims. Use representative tasks and failure cases. Compare demonstrations, documentation and contractual commitments. Where expert review is needed, identify the owner and the question they must answer.
  5. Decide and track conditions. Record approved scope, restrictions, unresolved actions and an accountable approver. Feed the result into procurement and the inventory; reassess when the use or supplier changes.

Example: Supplier A says customer prompts are not used for training. The reviewer records the service and plan covered by the statement, a current supporting document and any relevant contract provision. If the statement covers a different service, the answer remains unresolved. This illustrative example shows evidence handling, not a conclusion about any named supplier.

Back to page sections ↑

An original AI vendor assessment question set

Use these as discussion prompts. For every response, ask what proves it, which product it covers and whether it is a current practice or a future commitment. These are our own prompts, not copied official questionnaire fields.

  • Product and dependencies: Which models, providers and subcontractors power the proposed service? What changes can occur during our contract?
  • Data use: What inputs, outputs and logs do you retain? Are they used for training or improvement, and how is our selected configuration enforced?
  • Access and location: Who can access our data, for what purpose and in which deployment locations?
  • Evaluation: What evidence supports performance for our language, users and task? What failure conditions are documented?
  • Security: What protections and incident processes apply to this product? Which evidence can our security reviewer inspect?
  • Human oversight: What can users review, override or stop? What permissions govern those controls?
  • Traceability: What records can we access to investigate an output or decision, and for how long?
  • Change management: How are changes announced, evaluated and, where available, reversed?
  • Support and incidents: Who responds to a material problem, within what agreed process and with what evidence?
  • Exit: What happens to our records, integrations and retained data when the service ends?
Suggested evidence-log fields
FieldWhat to record
ClaimA precise statement about the proposed product and use.
EvidenceDocument or test reference, date, scope and reviewer.
FindingSupported, partially supported, unresolved or not applicable, with a reason.
ActionQuestion to resolve, owner and due date.
DecisionApproved scope, restrictions, approver and review trigger.

Escalate a material gap instead of averaging it away. Your team should define which issues prevent approval and which can be managed through a limited pilot or contract condition. This page does not supply a universal passing score.

Back to page sections ↑

Sources and next steps

NIST AI RMF Playbook: Manage ↗ provides context for ongoing third-party AI risk review. UK AI Assurance Questionnaire ↗ provides an official question navigator; its questions should be selected for the actual review. The comparison advice, evidence-log structure and example on this page are AI Gov Library editorial work.

Set authority through the AI governance policy template and find complementary forms in the templates guide. If a review spans teams, define the decision and escalation roles before requesting more questionnaires.

Back to page sections ↑

COMPARE RESOURCES

Which format fits?

Use a question bank for discovery, a scoring workbook for side-by-side reviews, and official risk guidance for control expectations.

01 Scan our suggested use02 Check format and access03 Review at the original source