AI Vendor NDA Template - No-AI-Training Clause (Free)

What an AI vendor NDA is

An AI vendor NDA is a standard vendor NDA plus clauses about training, retention, and model use.

When you share data with a vendor whose product doesn't touch AI, standard confidentiality terms cover the gap. When the vendor's product processes, stores, or could train on what you share, those terms leave out the parts that matter most. 

An AI vendor NDA covers the same ground as any confidentiality agreement, then adds clauses specific to how AI vendors handle data. If you're already familiar with what goes into the full NDA template, the AI-vendor NDA template is that, plus clauses about training, retention, and model use.

What a "no-AI-training" clause does (and does not do)

A no-AI-training clause restricts the vendor from using your confidential information to train or fine-tune AI models. It's a contractual promise you can enforce.

Whether it holds depends on the vendor's legal jurisdiction, how the clause is worded, and whether you can demonstrate a breach. If the vendor is a large model provider with its own terms of service, the clause you negotiate may conflict with those terms, which is a reason to read them before signing. 

• What the clause does: it puts the vendor on record, gives you a basis to pursue a remedy if the restriction is broken, and tends to prompt vendors to confirm what their product actually does with inputs. 
• What it doesn't do: technically prevent data from being used if the vendor ignores the agreement.

The AI vendor NDA clause checklist

Before reading the list, two terms worth defining:

The five clauses that separate an AI vendor NDA from a standard one:

1. No-AI-training clause: Restricts the vendor from using your confidential information, such as inputs, outputs, or metadata to train, fine-tune, or benchmark any AI model. Name the scope explicitly: does it cover the vendor's internal models, third-party model providers, or both?
2. Data retention and deletion on termination: Specifies how long the vendor retains your data and what happens when the contract ends. Without this clause, a vendor may retain inputs or derived embeddings indefinitely under their standard data policy. Require written confirmation of deletion.
3. Sub-processor and model-provider disclosure: If the vendor routes your data through another AI API (OpenAI, Anthropic, Google, or others) that provider becomes a sub-processor. This clause requires the vendor to disclose which sub-processors they use, and restricts them from adding new ones without notifying you.
4. Output ownership: Clarifies who owns content the AI generates using your confidential information as input. Without this clause, ownership of derivative outputs can be ambiguous, particularly if the model provider's terms assert rights over outputs.
5. Model-use restrictions: Limits what the vendor can do with the model after your engagement ends, for example, restricts them from using a model fine-tuned on your data to serve other clients.

When you need an AI vendor NDA vs a standard one

The question is whether the vendor's product could process or train on what you share.

If the vendor tells you their product doesn't use AI, ask them to confirm it in the NDA. 

Create your AI vendor NDA in minutes

The fastest way to get this done is to generate your AI vendor NDA online. Answer a few questions about the parties and the data you're sharing, and it builds the document with the AI-specific clauses already in place. Here’s how it works:

1. Select the AI generator option
2. State the AI vendor NDA case, name the disclosing party and the receiving vendor
3. Define confidential information. Be specific about the type of data you're sharing
4. Review the no-training, deletion-on-termination clauses, and the confidentiality term
5. Download, review, and send for signature

The limits of an NDA with AI vendors

An NDA is one contractual layer. Pasting confidential information into public AI tools may expose it if the tool stores or trains on inputs, and an NDA with a vendor you haven't vetted doesn't change what their product does technically.

Pair the NDA with: vendor due diligence (read their data processing terms before you sign), data minimisation (share only what the vendor needs), and where available, enterprise or no-training tiers that come with stronger contractual and technical commitments.

For a broader look at when AI use can trigger NDA obligations, see does using ChatGPT violate an NDA.

If you're unsure whether the clauses in this template cover your situation, consult a qualified attorney.