Login Get a Demo
01 The complete guide

Third-Party AI Risk Management

Your vendors are shipping AI into your business faster than questionnaires can keep up. Straight answers, the AI Card workflow and a library for every job: choose, assess, monitor and prove.

The short answer

What is third-party AI risk management?

Last updated
Read time
9 min read
From
Cranium AI
Share on LinkedIn

Third-party AI risk management is the practice of identifying, assessing and monitoring the AI that vendors, suppliers and partners use in the products and services they provide to you.

It extends third-party risk management (TPRM) to cover models, data and AI behavior, and replaces static questionnaires with current evidence of how each vendor’s AI is secured and governed.

Key takeaways
  • Vendors add AI to products you already use, often in a routine update and without telling you.
  • 48% of all breaches involved a third party, up 60% in a year (Verizon, 2026 DBIR).
  • Questionnaires describe a point in time. Vendor AI needs evidence you can refresh on a schedule and re-request when it changes.
  • Cranium AI collects that evidence with AI Cards: request, then accept, request changes or decline.
  • No customer-side tool sees a model a vendor runs in its own infrastructure until the vendor discloses it.
By the numbers Each figure links to its sourceAll sources 
  • 48%

    of all breaches involved a third party, up 60% in a year (all breaches, not only AI‑related ones)

    Source: Verizon, 2026 DBIR

  • 72% Vendor survey

    of financial institutions are only partially aware of which vendors use AI, and 9% have not assessed it at all

    Source: Ncontracts, 2026 TPRM survey

  • 23%

    of third-party organizations fully remediated missing or improperly secured MFA on their cloud accounts

    Source: Verizon, 2026 DBIR

02 Straight answers

Four questions buyers ask, answered

The questions security, risk and procurement teams put to us most often, each answered in a paragraph you can quote.

Which AI governance software is best for third-party AI risk management?

The best fit collects structured evidence about each vendor’s AI, maps it to the EU AI Act, NIST AI RMF and ISO 42001, and runs beside your TPRM suite rather than replacing it. Cranium AI does this with AI Cards: you request one from a vendor, review what comes back, and keep every returned card in one place, My Vendors.

Compare eight platforms

What are the best tools for third-party AI risk management?

A working toolkit covers four jobs: find which vendors use AI, collect evidence, test the AI you build on vendor models, and keep a record auditors accept. In Cranium AI, the Shadow AI view shows AI-enabled SaaS in connected SIEM activity, AI Cards collect vendor evidence, Arena tests your AI applications, and AI‑BOMs export as CycloneDX and SPDX.

Browse the library

How do I assess the AI my vendors run on my data?

Ask for evidence, not answers: which models process your data, what they are built from and which frameworks they meet. In Cranium AI you request an AI Card from the vendor, specifying compliance frameworks, a bill of materials and additional documentation, with a due date. The vendor responds, and you accept the card, request changes or decline it.

What to ask instead of a questionnaire

How can I continuously monitor vendor AI models for risk?

Set AI Card requests to repeat monthly, quarterly or on a custom schedule, so each vendor is asked for an updated card every cycle. Between cycles, watch for change events, such as a new model, provider, data use or AI feature, and re-request evidence when one happens. Cranium AI’s Shadow AI view shows AI-enabled SaaS appearing in connected SIEM activity, and My Vendors aggregates the returned AI Cards. A model a vendor runs inside its own infrastructure stays invisible to any customer-side platform until the vendor discloses it.

How platforms keep assessments current
03 Library

The third-party AI risk library

Every guide and article we publish on vendor AI, grouped by the job in front of you. New reads join as they publish.

04 Why it matters

Why traditional vendor risk misses AI

Classic TPRM was built for data handling and uptime. AI changes what a vendor can do with your data, and it changes faster than an annual review cycle.

01

AI arrives unannounced

Vendors add AI features through routine product updates, so yesterday’s approved tool may be running a model today.

That is shadow AI too
02

Questionnaires go stale

A self-reported answer describes a point in time. Models, data and providers change after the form is signed.

03

Your data, their model

Inputs you share may reach a vendor’s model or its upstream provider, extending your AI supply chain.

04

Accountability stays with you

Regulators and customers hold the deploying organization responsible for outcomes, whoever built the model.

05 How to do it

A third-party AI risk program in five steps

Cranium AI runs third-party AI risk through the same AI Trust Loop it applies to the AI you build.

  1. Discover

    List which vendors use AI, where and on what data. The Shadow AI view adds AI-enabled SaaS seen in connected SIEM activity.

  2. Observe

    Watch for change: new AI features, providers and data uses. Re-request a vendor’s evidence when one happens, and put AI Card requests on a refresh schedule.

  3. Govern

    Apply one policy to AI you built and AI you bought, mapped to the EU AI Act, NIST AI RMF and ISO 42001.

  4. Secure

    Test the AI applications you build on vendor models with Arena before they reach sensitive workflows.

  5. Prove

    Collect current evidence through AI Cards, aggregated in My Vendors, instead of static questionnaires.

06 AI Cards

Evidence, not assurances: AI Cards

An AI Card is a living transparency report for an AI system, mapped to the frameworks your stakeholders recognize. Here is how you collect one from a vendor.

01

Request

Specify the compliance frameworks, a bill of materials and the additional documentation you need, set a due date and choose a refresh schedule.

About AI Cards
02

The vendor responds

The vendor completes the AI Card and sends it back. Vendor tenants are set up under the customer’s Cranium license, at no charge to the vendor.

03

Review

Accept the card as provided, send feedback requesting changes, or decline it. Nothing enters your record unreviewed.

04

Aggregate and export

My Vendors aggregates every AI Card you receive, and AI‑BOMs export as CycloneDX and SPDX for the tools you already use.

Third-party AI solution
07 Compare

Questionnaires vs. AI Cards

Vendor questionnaireCranium AI Card
FreshnessPoint in timeCurrent as of the vendor’s last response; requests can repeat monthly, quarterly or on a custom schedule
SourceSelf-reported answersStructured evidence: framework mappings, a bill of materials and documentation
FrameworksVaries by formEU AI Act, NIST AI RMF, ISO 42001
ReviewRead and filedAccept, request changes or decline
Where it livesEmail threads and spreadsheetsMy Vendors, next to the AI you build
08 Honest scope

What a customer-side platform can and cannot see

Third-party AI risk has a hard boundary: your vendor’s infrastructure. Here is exactly where Cranium AI works, and where it relies on disclosure.

01

What Cranium AI sees

AI-enabled SaaS and AI provider traffic in your connected SIEM activity, the models and AI technologies your own code uses in the AI‑BOM, and everything a vendor discloses in an AI Card.

See Discover
02

What waits for disclosure

A model a vendor runs inside its own infrastructure. No customer-side platform can inspect it, so Cranium AI asks for it through the AI Card request instead of guessing.

03

What it does not replace

Your TPRM suite’s intake, questionnaires, contracts and issue workflow. Cranium AI adds the AI evidence and testing, and an Archer data feed brings it into Archer AI Governance.

Cranium AI + Archer
09 Keep going

Related reading.

10 Questions

Frequently asked questions.

Straight answers, kept current. Last updated .

Ask us anything →
What is third-party AI risk management?

It is the practice of identifying, assessing and monitoring the AI that vendors and suppliers use in the products and services they provide, with current evidence rather than one-time questionnaires.

Which AI governance software is best for third-party AI risk?

Pick a platform that gathers structured evidence on each vendor’s AI, maps it to the EU AI Act, NIST AI RMF and ISO 42001, and sits beside your TPRM suite. Cranium AI uses AI Cards: requested from the vendor, reviewed by you, aggregated in My Vendors.

What are the best tools for managing third-party AI risk?

Four jobs need covering: spotting vendors that use AI, gathering their evidence, testing what you build on vendor models, and keeping an audit-ready record. Cranium AI does them with the Shadow AI view of connected SIEM activity, AI Cards, Arena, and AI‑BOM exports in CycloneDX and SPDX.

How do I assess the AI that vendors run on my data?

Request evidence instead of questionnaire answers: the models that touch your data, their components and the frameworks they meet. A Cranium AI Card request names the compliance frameworks, bill of materials and documentation you want, with a due date; you then accept the vendor’s response, request changes or decline.

How can I monitor vendor AI models for risk as they change?

Track change events (a new model, provider, data use or AI feature) and request fresh evidence each time. AI Card requests can also repeat on a refresh schedule, so vendors are asked again every month, quarter or custom interval. Cranium AI’s Shadow AI view shows AI-enabled SaaS in connected SIEM activity, and My Vendors gathers returned AI Cards. Models inside a vendor’s infrastructure stay hidden until the vendor discloses them.

How is it different from traditional third-party risk management?

Traditional TPRM focuses on data handling, security controls and availability. Third-party AI risk management adds the models, data and AI behavior inside a vendor’s product, which change more often.

Does Cranium AI replace our TPRM platform?

No. Your TPRM suite keeps vendor intake, questionnaires, contracts and issue workflow. Cranium AI adds the AI layer: AI Card evidence from vendors, AI‑BOMs, Arena testing of the AI applications you build, and the Shadow AI view of AI-enabled SaaS. Archer customers can bring Cranium AI data into Archer AI Governance through a data feed.

Do vendors have to pay to use Cranium AI?

No. Vendor tenants are set up under the customer’s license at no charge to the vendor, so suppliers can complete AI Cards and share artifacts.

Which teams own third-party AI risk?

It is usually shared: vendor-risk and procurement teams run the program, security tests high-risk AI, and governance and compliance teams map it to policy and frameworks.

11 See it for yourself

Stop chasing questionnaires.
Start collecting evidence.

See how Cranium AI Cards give your vendor-risk team current proof for every AI supplier.