AI Considerations in Software Agreements: Key Clauses for the Age of Generative AI

Software-related agreements, such as licenses and development agreements, are increasingly considering artificial intelligence (AI) issues. Those that don’t should be.

These issues lie in several areas, including:

• The use of AI tools to develop software
• The use of AI tools to power software-as-a-service (SaaS) platforms
• Incorporating third-party AI tools into software
• Using unlicensed third-party content to train AI-based software
• Using client data on a SaaS platform to develop or refine a provider’s AI offerings

Specifically, concerns often focus on Generative AI (GAI) tools like ChatGPT.

As the Copyright Office explains,

These technologies “train” on vast quantities of preexisting human-authored works and use inferences from that training to generate new content. Some systems operate in response to a user’s textual instruction, called a “prompt.” The resulting output may be textual, visual, or audio, and is determined by the AI based on its design and the material it has been trained on.

For example, a GAI tool can use a prompt to generate “new” software code to perform a specific task stated in the prompt.

GAI can also be used to summarize news stories and other documents and to “answer questions.”

However, there are several issues associated with GAI, including:

• As we’ve reported in previous blogs, many copyright owners have sued GAI companies because they object to their content being used, without their permission, to train the GAI tools. Courts have not yet determined whether this training is allowed as “fair use” under copyright law.
• GAI tools are known to “hallucinate”: they can state “facts” that are wrong and offer “advice” or “analysis” that’s wrong or even dangerous.
• GAI tools use an inordinate amount of water and electricity in proportion to the value they generate.

Because of such issues, some clients have started to include language like the following in their software development and licensing agreements:

Generative AI.

(a) The Company shall not use any generative artificial intelligence tools, large language models, or similar technologies (“AI”) (including any licensed from or otherwise made available by a third party (“AI Provider”)) in performing the Services without the prior written consent of Client which may be granted or withheld in Client’s sole discretion (each, an “AI Consent”). To the extent Client provides an AI Consent, Client’s AI Consent is limited to the AI Provider(s), AI technology, and Services described in such consent. Company agrees to notify Client in writing if, during the term of this Agreement, Company intends to change or add an AI Provider with respect to any Services or the AI technology used by such AI Providers. Client reserves the right to review and approve the use of any AI Provider before such AI Provider’s AI technologies are implemented in the Services covered by this Agreement.

(b) The Company shall provide prior notice to Client of any features added to the Services related to AI technology and shall reasonably cooperate with Client requests for supplemental information regarding such features, including by providing the following information:

• 1. the name of the AI and (if applicable) the AI Provider;
  2. a description of the manner in which the AI will be used in performing the Services, including the output of such AI that will be used (“AI Output”);
  3. a description of any Client Data, or other Client Confidential Information, that the Company proposes to provide as an input to the AI (including a description of any Client Personal Data that may be included in such input) (“AI Input”);
  4. a description of (a) (in the case of AI provided by an AI Provider) any access that the AI Provider has to AI Inputs, if any, and (b) the use of AI Inputs for the model’s training set, if any, including (in the case of AI provided by an AI Provider) whether Company has any contractual agreements or terms of use with the AI Provider that restrict the ability of employees or contractors of the AI Provider from viewing AI Inputs and/or that prevent AI Inputs from being used to train any model that is used by other users; and
  5. a summary of any diligence conducted by the Company to confirm that the AI Output is accurate, unbiased, and explainable.

Such clauses are similar to common clauses in open-source code, like the following example.

No Open Source. The developer warrants that the Software does not contain any Open Source Software unless it is disclosed to and approved by the Customer prior to being included.

Software development agreements and licenses also commonly include language stating that no third-party rights are being infringed by the software. A simple version of such a clause is:

No Third Party Property. The developer represents and warrants that it will not use any trade secrets or confidential or proprietary information owned by any third party in developing the Software.

It’s especially important to include such clauses when a GAI tool has been trained on potentially confidential or proprietary information owned by third parties.

A client may want to prevent its own data from being used to train the AI (which could then potentially be provided to its competitors) by including language like the following:

All AI Input provided by the Client (“Client AI Input”) and all AI Output generated by the AI for the Client (“Client AI Output”) shall be owned by the Client. All Client AI Input and Client AI Output shall be deemed to be Client Confidential Information. Company and any Company Agents shall use any Client AI Input and Client AI Output solely in connection with the Services and shall not provide it to any third party except as required to provide the Services.

Company and any Company Agents shall not use any Client Data or other Client Confidential Information (including any Client AI Input or Client AI Output) to train any AI, even if deidentified, aggregated, or anonymized.

It’s common for a software development agreement to warrant that the software to be developed won’t infringe any third-party rights. For example, a simple version might be something like:

No Infringement. Developer represents and warrants that the Software will not infringe any patent, copyright, trademark, trade secret or other proprietary right of any person.

Again, it’s important from the client/user perspective for a software developer to include such a warranty when a GAI tool has been trained on unlicensed third-party content.

Finally, from a client’s perspective, it’s important to be indemnified against any breach of warranties related to AI. A simple version of such a clause is:

Indemnification. If any third party brings a lawsuit or proceeding against Customer based upon a claim that the Software breaches the third party's patent, copyright, or trade secrets rights, then Developer shall hold Customer harmless against any loss, damage, expense, or cost, including reasonable attorney fees, arising from the claim.

Of course, to be worth anything, an indemnification clause needs to be enforceable, and the indemnifying party (in this case, the developer) needs to have the resources to make good on a promise to indemnify. A client/licensee would be wise to ensure that a software developer has insurance to cover indemnification obligations.