Skip to main content

AI Vendor Risk Manager

Risk, audit and assurance, a mid-level role

What does AI Vendor Risk Manager do?

Evaluates the risk of buying, licensing, embedding or relying on external AI: data use, security, model changes, subcontractors, IP, availability, audit rights and accountability the customer cannot outsource.

What it decides: The risk tier of a vendor and use, what evidence is required, and whether to approve, condition, pilot, defer or reject.

The competencies employers name

  • AI vendor due diligence and third-party riskcore, depth expected

    Tiers vendors by use and impact, requests evidence instead of promises, tests in the customer's context, and plans monitoring and exit.

    5 graded topics teach this

  • Contract terms that allocate AI riskcore, depth expected

    Turns controls into enforceable obligations: data use, change notice, audit rights, incident duties, subcontractors, IP, exit and deletion.

    4 graded topics teach this

  • AI risk and impact assessmentrequired, working knowledge

    Reviews purpose, data, affected people, accuracy, bias, security, oversight, vendors and law for a use case, scores likelihood and impact, and documents residual risk.

    14 graded topics teach this

  • AI resilience, continuity and exit planningrequired, working knowledge

    Plans for a vendor failure, an unsafe model change or a suspended service: rollback, manual fallback, data export and safe decommissioning.

    3 graded topics teach this

  • Privacy law applied to AIrequired, working knowledge

    Applies GDPR, CCPA and sector rules to training data, inference, automated decisions, lawful basis, individual rights and cross-border transfer.

    7 graded topics teach this

  • AI security fundamentalsrequired, working knowledge

    Understands prompt injection, data poisoning, model theft, insecure integrations and excessive agent privileges, and the controls that reduce each.

    17 graded topics teach this

  • Buying AI wellrequired, working knowledge

    Writes requirements, runs a fair evaluation, pilots within limits, and refuses a demo as evidence.

    15 graded topics teach this

  • Post-deployment monitoring and drift detectionpreferred, working knowledge

    Sets performance metrics, thresholds and review triggers after launch, and treats a model change, a vendor update or new data as a reason to re-check.

    12 graded topics teach this

  • Executive and board communication on AI riskpreferred, working knowledge

    Turns technical uncertainty into a one-page decision: material risks, trends, exceptions, remediation, and what the board is being asked to accept.

    10 graded topics teach this

  • Cross-functional facilitation and influencerequired, working knowledge

    Interviews, facilitates, challenges and secures action across legal, security, product and business teams without owning every decision.

    20 graded topics teach this

Where it is taught

Counted from the graded topics that teach this role's competencies. Your own path is shorter: it skips what you already cover.

Check your readiness for this role

Add what you already have (optional)
Signed in? Every topic you have passed already counts as proof.

Roles that feed into it

  • Third-Party Risk Analyst
  • Procurement Manager
  • Vendor Manager
  • Security Assessor
  • Privacy Analyst

Where it leads

  • Director of Third-Party Risk
  • AI Risk Director
  • Chief Procurement Officer

Backgrounds that reach it fastest

What postings tend to name

Frameworks: NIST AI RMF GOVERN 6 and MANAGE 3, ISO/IEC 42001, CISA software acquisition guidance.

Credentials often listed: CTPRP, CRISC, CGRC, CIPP, AIGP. GAGE does not issue these and does not prepare for their exams; the record you earn here is your own graded evidence, which stands beside them.

Questions

Why is an ordinary vendor review not enough for AI?
Performance varies by context, outputs are probabilistic, the model can change after approval, explanations may be unavailable, data rights can be unclear, and the supply chain is layered. A questionnaire built for software does not reach those.
CallEmailHelp