Skip to main content

    Vendor Risk

    AI Vendor Risk Assessment: What to Check Before Buying

    In short

    Decide which AI vendor answers need evidence before you buy, what evidence to ask for, and how to record the decision.

    a supplier review: permissions matrix, data-retention calendar and assurance report folder.

    Company names and figures are invented.

    Before an AI vendor gets access to your data or systems, connect its answers to records you can review. This guide maps questions about data, access and model changes to evidence and a purchasing decision.

    Look beyond a completed questionnaire

    An AI vendor questionnaire records what the supplier says. Before sharing confidential data or allowing an agent to act in business systems, you need to know what supports the important answers.

    Treat questionnaire answers as self-reported. For higher-risk uses, reviewers inspect contracts, settings and assurance records, and test selected controls where appropriate. The NIST AI RMF helps organize third-party risks and responsibilities.

    Evidence for vendor risk decisions.

    ExposureData + permitted actions
    EvidenceRecords / control tests
    Owner decides
    Accept
    Restrict
    Defer
    Example

    Identify what data enters and what actions the vendor can perform. Trace retention, access and model-change claims to records or control tests. Document unresolved evidence needs when deciding whether to accept, restrict or defer use. Evidence depth follows risk; unknown is not the same as satisfactory.

    Questions to ask before you buy

    Reference table · every row is included.

    AI vendor questions and useful supporting records
    AreaAsk the supplierLook for
    Data useCan our inputs or outputs be used for training?Contract terms and applicable tenant settings.
    RetentionHow long is our data kept and how is it deleted?A retention schedule and deletion process, including backups.
    AccessWho can access our data and what can connected tools do?Roles, administrative access arrangements and integration permissions.
    Other providersWhich providers process our information, and where?Subprocessor lists, locations and contractual responsibilities.
    ChangesHow will we learn about changes that affect our use?Change notices, version policies and customer choices.
    IncidentsHow will you notify us and help investigate?Notification commitments and response procedures.
    AssuranceDoes your assurance report cover the product we will use?The actual report, its coverage, dates, exceptions and customer responsibilities.
    ExitCan we export our records and close the service safely?Export options, deletion commitments and termination terms.

    Example: reviewing a contract assistant

    A SOC 2 report may help, but its product coverage, period, exceptions and customer responsibilities still need review. It does not certify the accuracy of model outputs or every action an agent can take. The AICPA SOC resources explain these reports.

    Record the decision, unresolved questions and the person accepting any remaining risk. Later reviews can then check whether the conditions were met. See ISO 42001 audit preparation.

    Match the review to the risk

    A tool used only for public information may need less work than one handling client files, medical data or payments. Consider the sensitivity of information, actions the system can take, dependence on the supplier and the difficulty of reversing a mistake.

    Our AI risk and controls services help examine these concerns. Record suppliers and decisions in the AI system inventory, and use the AI Governance QuickScan for broader ownership and policy questions.

    Key takeaways

    • Ask what supports the answers that matter most.
    • Check coverage and exceptions before relying on a supplier report.
    • Record unresolved questions and review decisions after material changes.
    Related serviceAI Technical Controls AssessmentRelated resourceAI System Inventory TemplateFree toolAI Governance QuickScan

    References

    Primary sources cited above. Standards are referenced by number and year; their text is licensed and is paraphrased, not reproduced.

    1. NIST AI 100-1, AI Risk Management Framework (AI RMF 1.0), January 2023 (opens in a new tab) — Map function, including third-party components and responsibilities.
    2. NIST AI 600-1, AI RMF: Generative Artificial Intelligence Profile, July 2024 (opens in a new tab) — Risk categories relevant to evaluating model changes and supplier models.
    3. NIST AI Risk Management Framework program page (opens in a new tab)
    4. AICPA, 2017 Trust Services Criteria with Revised Points of Focus (2022) (opens in a new tab) — Criteria a SOC 2 examination is performed against.
    5. AICPA, System and Organization Controls (SOC) suite of services (opens in a new tab)
    6. ISO/IEC 42001:2023, Artificial intelligence management system — Requirements (opens in a new tab) — Licensed text; supplier-related control objectives referenced, not reproduced.

    The next step

    Discuss a professional assessment.

    Check how a supplier handles your data, access, changes and incidents before you commit.