[ Docs ](https://sendnda.com/docs)     

 [ The agreement ](https://sendnda.com/docs/the-agreement)     

Use of AI tools
===============

Three levels for what the receiving party may do with AI tools, and why the default sits in the middle.

 02

The agreement
-------------

Article 3 of 5

 Most teams paste things into an AI tool now: a deck to summarise, a spreadsheet to check, a contract to explain. An NDA that says nothing about it leaves both sides guessing. Every Send NDA agreement says what the Receiving Party may do with AI tools, and you pick one of three levels.

Whatever you pick, one line never moves: the Receiving Party may not use your information to train, fine-tune, adapt or otherwise improve a model.

Three levels
------------

 | On the form | What the Receiving Party may do |
|---|---|
| No AI tools | Use a service only if it keeps nothing after answering, doesn't train on the input and doesn't pass it on. |
| AI tools that do not train on the information | Use a service whose terms rule out training on the input and sharing it, and let the provider keep it only as long as it needs to run and monitor the service. |
| Any AI tool | Use any service for the purpose. They still may not train a model on it themselves. |

Why the middle is the default
-----------------------------

The strictest level reads well and is hard to live with. Few services keep nothing at all: most keep input for a while to watch for abuse, so under "No AI tools" the Receiving Party can use almost none of them. Pick it for information that shouldn't leave the building.

The middle level matches how business and API plans are sold: no training on your input, no sharing beyond the provider's own subprocessors, retention only for running the service and watching it for security and abuse. That's where most teams working with AI already are, so it's the default. The wording:

> The Receiving Party may make Confidential Information available to an artificial intelligence or machine learning service for the Purpose only where the terms on which that service is provided exclude use of the input to train or improve any model, exclude disclosure of the input to any third party other than the provider's subprocessors bound by the same terms, and allow the provider to retain the input only for as long as is needed to provide the service and to monitor it for security and abuse.

The open level
--------------

"Any AI tool" lets the Receiving Party use whatever they like, consumer apps included. What they may not do is train a model on your information themselves, or put it in a dataset for that. It doesn't stop a provider from doing what its own terms allow, which is the difference with the middle level. Pick it when the information is confidential but not sensitive, and friction matters more than the last bit of control.

Naming a service
----------------

When the Receiving Party is a company, the contract adds one more sentence at every level: you may tell them in writing not to use a named service for your information, and they have to stop. It's the way to react when a provider changes its terms mid-deal. When the Receiving Party is a person, the sentence is left out.

Older agreements
----------------

An agreement made on a template before version 2.2 carries the "No AI tools" wording. [The optional clauses](https://sendnda.com/docs/the-agreement/clauses) covers the rest of what you can switch on.

These are summaries to help you choose. The contract's own wording binds, and Send NDA provides a template, not legal advice.

 Checked against the contract template on 21 September 2026.

 [   The optional clauses ](https://sendnda.com/docs/the-agreement/clauses) [ How long it runs   ](https://sendnda.com/docs/the-agreement/how-long-it-runs)