AI Subprocessor

What is an AI subprocessor?

An AI subprocessor is an outside AI company that one of your vendors uses to process your data. When your help desk tool summarizes tickets or your CRM drafts emails, the AI behind it often runs on another company's servers.

A subprocessor is any company your vendor hires to handle personal data on your behalf. An AI subprocessor is the same idea, applied to AI providers. Names like OpenAI and Anthropic now appear on vendors' subprocessor lists.

Under the GDPR, a vendor that processes personal data for you needs your authorization before it adds another processor. Article 28(2) allows "general written authorisation," as long as the vendor tells you about new additions so you can object. Article 28(4) says the new processor must take on the same data protection duties as your vendor.

How your data reaches an AI subprocessor

1. You or your team use an AI feature inside a vendor's product.
2. The vendor sends your text, files or records to a model provider to get an answer.
3. The model provider may run the model itself or through a cloud platform it works with.
4. The answer comes back to the vendor, which shows it to you.

At each step, someone may log, store or review the data. Your contract covers only the vendor. The rest depends on agreements you never see.

What it looks like in a vendor's terms

A typical clause reads like this:

We use third-party AI service providers, listed on our subprocessor page, to provide AI features. These providers process Inputs and Outputs only to deliver the features and may not use them to train their models. You agree to follow each provider's usage policies.

• "Listed on our subprocessor page" means the list, not the contract, names the providers. The list can change.
• "Only to deliver the features" limits what the provider may do with your data.
• "May not use them to train" is a no-training promise at the provider level. Compare it with any zero data retention promise, which is about storage.
• "Follow each provider's usage policies" adds the AI provider's rules to your obligations, even though you did not sign with that provider.

Mixpanel's Supplemental Terms include a clause of this kind. As Venpo reported in June 2026, they point to the AI providers on Mixpanel's subprocessor list and require customers to follow those providers' usage policies.

Model Provider (LLM Provider)

A model provider, also called an LLM provider, is the company that builds and runs the large language model behind an AI feature. A large language model (LLM) is the type of AI that reads and writes text. Your vendor builds the feature and the screen you see. The model provider supplies the "brain" that produces the answers.

A vendor may not build its own model at all. It can pay a model provider per request, through an API, which is a way for one system to send data to another.

The same model can often be reached in more than one way. Vercel's AI Gateway documentation, for example, lists Anthropic models as available directly from Anthropic and through Amazon Bedrock and Google Vertex AI. Each route can come with its own data terms.

When a vendor launches an AI feature, ask three questions. Which model provider runs it? Through which platform? And which terms govern your data at that provider?

A real change we caught

Venpo flagged that RingCentral adds AI subprocessors almost every month, and that at least 30 vendors added Anthropic as a subprocessor since September. We also caught a case where SurveyMonkey changed its AI providers on a page still dated November 2025, which shows why the date on a list is not enough. For a practical method, read our guide on how to monitor AI subprocessors in your vendor stack.

Why it matters to you

Your data travels further than your contract. Each AI subprocessor is a fourth party: a company you have no contract with that still handles your data.

Your customers may need to know. If you pass your customers' personal data to vendors, your own subprocessor list and DPA may need to reflect the AI providers behind them. New ones can trigger notice duties toward your customers.

Location and retention can change. A new AI provider may process data in another country or keep it for misuse checks. That can affect transfer rules and what you told customers about retention.

Objection windows are short. If your DPA gives you a right to object to a new subprocessor, it may run for only a set number of days after notice. Missing the notice means missing the window.

What to check

• Read each vendor's subprocessor list for AI providers, and note which features they power.
• Check whether the vendor's DPA requires notice before adding a subprocessor, and how you receive it.
• Ask whether each AI provider is bound by no-training and retention limits for your data.
• Check where each AI provider processes data, and which transfer mechanism covers it.
• Ask whether you can turn off AI features that rely on a provider you do not accept.

How Venpo helps

Venpo watches your vendors' subprocessor lists and DPAs and emails you when an AI provider is added, removed or given a new role, with what changed and why it matters. Each vendor's change history shows when every provider first appeared.