Model providers
The company whose model answers the prompt. Your vendor calls its API or hosts its weights.
Why your vendor’s model provider is your fourth party, what DORA and the GDPR say about subcontracting chains, and how to manage the exposure.
Fourth-party AI risk is the exposure an organization takes on from the suppliers its own vendors rely on to deliver AI: model providers, cloud and compute hosts, data sources and other subcontractors.
You hold no contract with these parties, yet your data may reach them, and their failures reach you through the vendor.
Most AI features are assembled, not built. Each layer below your vendor is a party you depend on without a contract.
The company whose model answers the prompt. Your vendor calls its API or hosts its weights.
Where the model runs. The FSB notes that financial institutions increasingly take on indirect cloud exposure through specialised AI service providers.
FSB report, 2024Retrieval sources, plug-ins and agent tools the vendor wires in, each with its own operator.
Any party below the vendor that processes your personal data, which data protection law treats separately.
AI subprocessor| Rule | What it says |
|---|---|
| DORA, Article 29(2) | Where a provider may subcontract services supporting critical or important functions, weigh the benefits and risks, and assess whether long or complex chains affect your ability to monitor the function |
| DORA, Article 30(2)(a) | The contract states whether subcontracting of such a service is permitted, and on what conditions |
| Delegated Regulation (EU) 2025/532 | In force since July 22, 2025. The contract makes the provider responsible for subcontracted services, requires it to monitor them and gives you notice of material changes in time to assess the risk |
| GDPR, Article 28(4) | A processor that engages another processor imposes the same data protection obligations by contract, and stays fully liable if that processor fails |
| US interagency guidance (2023) | Due diligence weighs how far a third party relies on subcontractors; contracts can require notice of their use |
Have each vendor name the model providers, hosts and subprocessors behind every AI feature.
Ask for an AI bill of materials or an equivalent disclosure, so the list is a document and not a sales answer.
Require the vendor to bind its suppliers to your data-use, security and notice terms.
A swapped model provider is a material change. Ask for notice before it happens, with a right to object.
Record which fourth parties recur across vendors. Shared dependencies are concentration risk.
Straight answers, kept current. Last updated .
Ask us anything →Third-party AI risk comes from a vendor you contract with directly. Fourth-party AI risk comes from that vendor’s own suppliers, such as the model provider behind its AI feature. You can negotiate with a third party. With a fourth party, your only levers are the terms you require your vendor to pass down and the evidence you require it to produce.
Often both. Fourth party is a risk term for any supplier your vendor depends on. Subprocessor is a data protection term for a party that processes personal data on your behalf under your vendor. A model provider that receives prompts containing personal data can be both; one that never sees personal data is a fourth party only.
Only what a vendor discloses or what your own code reveals. Cranium AI collects vendor disclosures through AI Cards, where you can request a bill of materials, and generates AI‑BOMs from your code repositories that list the models and technologies your applications call. A model a vendor runs in its own infrastructure stays invisible until the vendor discloses it.
Primary sources for the figures and frameworks cited on this page. Last verified .
See how Cranium AI discovers, observes, governs, secures and proves every AI model, agent and vendor in your enterprise.
Cookie Policy · Privacy Policy
Global Privacy Control detected. Opt-out request honored: marketing tags stay off in this browser.