River
Y CombinatorBacked by Y Combinator
Free document

AI Addendum

Common Paper's standard AI addendum, version 1.0: cover page choices on training with customer data, output ownership, indemnity and AI acceptable use.

Free download  ·  No account needed

Want it written around your situation? Describe it in the Founder Document Studio for Startups.

About this document

An AI addendum is a short contract add-on that sets the rules for the AI features in a software product. It covers what data the provider may train on, who owns what the AI produces, and what the provider does not promise about it. This is Common Paper's AI Addendum, version 1.0, shown word for word as Common Paper publishes it.

Add the AI Addendum to a Cloud Service Agreement or Software License Agreement already signed on Common Paper's standards. If your contract is not built on those standards, its defined terms will not line up. By default the provider may not train AI models on the customer's inputs or outputs unless the cover page allows it, so decide the training choices before you send it.

How to fill it in

  1. Name the signed agreement. Describe the existing Cloud Service Agreement or Software License Agreement this AI Addendum changes, including the parties and its date.
  2. Choose the training terms. Tick which data the provider may train on, if any, for whose benefit, and any limits such as removing details that identify people. Gray drafting notes explain each choice.
  3. Set covered claims and use rules. Decide whether the indemnity (who pays if a third party sues) covers claims about the AI's output or the customer's input, and attach any AI acceptable use policy.
  4. Delete the notes and sign. Remove the drafting notes and unused options, then have a person authorized at each company sign the cover page.
AI Addendum

COVER PAGE

AI Addendum

[Drafting note: This AI Addendum is intended to be added to an existing, already signed Cloud Service Agreement or Software License Agreement using the Common Paper standards. Fill out the yellow highlighted field below with details about the existing agreement. For example, "Cloud Service Agreement between Rose Apothecaries, Inc. and Vandelay Industries, Inc. dated January 1, 2025".]

This Cover Page incorporates the AI Addendum Standard Terms available at https://commonpaper.com/standards/ai-addendum/1.0/ with the Variables set forth below (collectively, the "AI Addendum"). A copy of the AI Addendum Standard Terms is attached for convenience only.

This AI Addendum amends and is incorporated into the following "Agreement": [ insert description of Agreement ]. Undefined capitalized words have the meanings or descriptions given in the Agreement. If there is any inconsistency between this AI Addendum and the Agreement, the AI Addendum will control for the provision of AI Services.

Training Data

[Drafting note: By default, the provider cannot train any models using customer content, information, or data unless authorized below. Select the categories of data or information that may be used to train models. If no model training is permitted, check "None" and delete the rest.]

[ ] None

Provider may Train the Model(s) using the following Training Data:

[ ] Usage Data

[ ] Feedback

[ ] Input

[ ] Output

[ ] User prompts, excluding other components of Input

[ ] Customer Content

[ ] [ free text box ]

Training Purposes

Permitted Model Training

[Drafting note: If model training is permitted, select the purposes for which the trained models may be used. If no model training is permitted, check "None" and delete the rest.]

[ ] None

Provider may use Training Data for the following purpose:

[ ] Train the Model(s) in the AI Services solely for Customer’s benefit

[ ] Train the Model(s) in the AI Services

Training Restrictions

Restrictions on Model Training

[Drafting note: This is an optional variable. If there are specific restrictions on what must be done before training may be performed, specify that here. Otherwise, write "None" or delete the row in its entirety.]

[ ] Training Data must be aggregated

[ ] Training Data must be de-identified

[ ] Provider will use commercially reasonable efforts consistent with industry standard technology to de-identify Training Data

[ ] [ free text box ]

Improvement Restrictions

For improvements to the AI System (but not any Models)

[Drafting note: By default, the provider can use input, output, and training data (if any are defined) to make general product improvements that are not model training. Use this section to add protections and restrictions around the general product improvement. If there are none, write "None" or delete the row in its entirety.]

[ ] Neither Input nor Output may identify Customer

[ ] Improvements based on Customer's Input, Output, or Training Data will be solely for Customer's benefit

[ ] [ free text box ]

Covered Claims

Claims covered by indemnity obligations

[Drafting note: This AI Addendum leverages the common language across Common Paper's CSA and Software License Agreement. Use this section to leverage the existing structure in those agreements to add indemnification obligations specific to the AI Addendum. Update and modify the language as needed. If there are no indemnities specific to the AI Addendum, delete this row in its entirety.]

[ ] Provider Covered Claims include any action, proceeding, or claim [ that the Output—when generated and used by Customer according to the terms of the Agreement and the AI Addendum—violates, misappropriates, or otherwise infringes upon the intellectual property or other proprietary rights of another person or entity.

Without limiting the indemnity exclusions in the Agreement, Provider's obligations as an Indemnifying Party will not apply to Provider Covered Claims that result from: (a) use of Output in combination with data, software, hardware, equipment, technology, or other products or services not provided by Provider; (b) Input; (c) Customer's use of the AI Services in breach of the Agreement or the AI Addendum; (d) modifications to the Output that were not made by Provider; (e) Output that Customer knew or should have known might violate, misappropriate, or otherwise infringe upon another's intellectual property or other proprietary rights; or (f) a claim that use of Output infringes another's trademark or related rights. ]

[ ] Customer Covered Claims include any action, proceeding, or claim [ that (1) the Input—when used by Provider according to the terms of the Agreement and the AI Addendum—violates, misappropriates, or otherwise infringes upon the intellectual property or other proprietary rights of another person or entity; or (2) results from Customer's use of the AI Services in violation of the applicable restrictions in the Agreement or the AI Addendum. ]

AI Acceptable Use Policy

[Drafting note: Many AI products, particularly generative AI products, include an acceptable use policy specific to how users are permitted to interact with the AI features. Attach or link to yours here.]

[ ] Use of the AI Services is subject to the Acceptable Use Policy [ available at URL or attached to this Cover Page ]

By signing this Cover Page, each party agrees to enter into this AI Addendum.

PROVIDER: [official company name]

CUSTOMER: [official company name]

Signature

Print Name

Title

Notice Address

Use email or postal address

Date

STANDARD TERMS

AI Addendum Standard Terms

1. AI Services

1.1 Using AI Services. The AI Services are part of the Product and subject to the Agreement as supplemented by this AI Addendum. Customer may use AI Services by providing Input. The AI Services may generate Output in response to Input. Provider may copy, display, modify, distribute, and use Input to the extent necessary to provide the AI Services as contemplated by this AI Addendum. Customer authorizes Provider to process Input for all such purposes.

1.2 Restrictions. Without limiting the restrictions contained in the Agreement, Customer will not (and will not allow anyone else to): (a) use the AI Services for decision-making in a regulated industry or capacity without proper human oversight and review in compliance with Applicable Laws and applicable professional ethics, guidelines, and rules; (b) use the AI Services to violate, misappropriate, or otherwise infringe the intellectual property or other proprietary rights of others; or (c) falsely state Output was created by a human.

1.3 Model Training. Unless the Cover Page identifies Training Data and Training Purposes, Provider may not use Customer's Inputs or Outputs to Train any Model. If the Cover Page permits Training, then subject to the Training Restrictions, Provider may copy, modify, distribute, and use Training Data for the Training Purposes.

1.4 Non-Training Improvement. Subject to the Improvement Restrictions, Provider may use Input, Output, and Training Data to provide, maintain, develop, and improve the AI System, provided that such usage does not constitute Training except to the extent authorized for Training Purposes.

2. Intellectual Property and Privacy

2.1 Ownership. As between the parties, Customer (a) retains all right, title, and interest in and to all Input, and (b) owns all Output. To the extent permitted by Applicable Laws, Provider hereby assigns to Customer all right, title, and interest—if any—in and to Output.

2.2 Personal Data. Nothing in this AI Addendum will reduce or limit Provider's obligations under Applicable Data Protection Laws regarding Personal Data that may be contained in Input.

2.3 Rights to Input. Customer represents and warrants that it, all Users, and anyone submitting Input each have and will continue to have all rights necessary to submit Input to the AI Services.

3. Disclaimers

3.1 Nature of AI. Due to the nature of artificial intelligence and machine learning, information generated by the AI Services may be incorrect or inaccurate. The AI Services are not human and are not a substitute for human oversight. Output generated by the AI Services may not be protectable as intellectual property.

3.2 Similarity of Output. Output may resemble or be duplicative of data, information, and materials created by the AI Services for others. Provider does not provide any representation or warranty that Output (a) does not and will not incorporate or reflect the data, information, prompts, or materials of others, (b) will not violate, misappropriate, or otherwise infringe upon the intellectual property or other proprietary rights of another person or entity, or (c) will not be reproduced in the same or similar way to another user of the AI Services.

4. Definitions

4.1 "AI Addendum Standard Terms" means these Common Paper AI Addendum Standard Terms Version 1.0, which are posted at https://commonpaper.com/standards/ai-addendum/1.0/.

4.2 "AI Services" means the artificial intelligence or machine learning components of the Product, including the AI System and underlying Model(s).

4.3 "AI System" means the artificial intelligence or machine learning application, program, and services layers of the AI Services, excluding the underlying Models.

4.4 "Input" means the data, information, prompts, or materials submitted by or on behalf of Customer or Users to the AI Services but excludes Feedback.

4.5 "Model" means a large language, machine learning, or artificial intelligence model.

4.6 "Output" means the data, information, or materials created by the AI Services in response to Input.

4.7 "Train" or "Training" means the use of data, information, or materials to create or improve a Model.


Common Paper AI Addendum (Version 1.0) free to use under CC BY 4.0.

Page 1 of 41,485 words

Frequently asked questions

Who owns AI output under the Common Paper AI Addendum?

The customer. The standard terms say the customer keeps all rights in its input and owns all output, and the provider assigns any rights it has in the output to the customer, where the law allows. The terms also warn that output may resemble what the AI produces for other customers.

Can a SaaS provider train AI on customer data?

Only if the contract allows it. Under the Common Paper AI Addendum, the provider may not train models on the customer's inputs or outputs unless the cover page names the training data and purposes. Without an AI addendum, check the main agreement's data use and confidentiality terms.

Does the AI Addendum promise the AI is accurate?

No. The standard terms state that AI output may be incorrect and is not a substitute for human oversight. The customer also agrees not to use the AI for decisions in a regulated field without proper human review.