Are you struggling to understand how AI vendor agreements differ from typical SaaS agreements? You’re not alone. As businesses rapidly adopt artificial intelligence tools, in-house counsel and legal professionals face unique challenges when reviewing AI contracts.
In a recent webinar hosted by Contract Nerds 📄🤓 on “How to Redline AI Vendor Agreements: Customer vs. Vendor Tactics for Key Clauses”, legal experts David Tollen, Laura Greenberg, and Nada Alnajafi shared practical strategies for reviewing and negotiating AI terms that balance both customer and vendor concerns. The speakers evaluated OpenAI’s Business Terms as part of their deeper dive into five commonly negotiated areas in AI vendor contracts.
With an audience of 913 live attendees and 1,887 registrants, this is surely a hot 🔥topic in the legal industry. Then again, the Contract Nerds webinar program always draws a large crowd!
Watch and Learn: Watch the full 75-minute webinar recording and access the presentation and bonus materials to dive deeper into this topic.
1. IP Indemnification for AI-Generated Outputs
When your company uses AI-generated content in business operations, who defends you if you’re sued for IP infringement? This question is crucial when negotiating AI contracts.
“If you’re on the customer side, don’t assume you’re going to get this [IP indemnity] from all of your AI vendors. OpenAI and a few other large language models have taken a big step in the last few years and reversed an earlier policy and they’ve started to give IP indemnities,” noted, David Tollen of Tech Contracts Academy.
However, these are some common IP indemnity exceptions in AI vendor agreements:
- No coverage if you knew the output might infringe
- No protection if you disabled safety features
- No trademark infringement coverage
- No protection if your input data contained the infringing material
What was interesting is that nearly 60% of people asked said that they found these indemnity exceptions reasonable and would leave them as-is. See results from the webinar poll below.
💡Pro tip: Most AI vendors offer limited indemnities but avoid warranties about outputs because of the unpredictable, uncontrollable nature of AI outputs. If you can’t get the warranties you’re looking for, remember that vendors may be more willing to offer indemnities as a business decision.
Read and Learn: Lawyers should be on the lookout for unique issues in contracts involving the purchase or sale or AI. Artificial Intelligence Contracts: An Issue-Spotter by David Tollen.
2. Warranties About AI Output Accuracy
As discussed above, standard AI contracts typically disclaim warranties about output accuracy. The unpredictable nature of generative AI means most vendors simply cannot guarantee perfect outputs. This represents a major difference from traditional software agreements where vendors make specific functionality promises.
The speakers discussed an interesting trend where vendors based on large models are less likely to offer output warranties while vendors based on smaller models are more likely to.
As you negotiate AI terms with more established vendors, consider pushing for the following accuracy guarantee alternatives:
- Warranties that outputs will meet documented specifications
- Warranties about the accuracy of the vendor’s known issues disclosures
- Service level agreements addressing output quality metrics
💡Pro tip: Ask the vendor how they measure accuracy and then focus your negotiation on specific, measurable aspects of output quality rather than broad warranties.
3. Ownership Rights for AI Inputs and Outputs
Most AI terms acknowledge that, as between vendor and customer, customers own their own inputs. But the treatment of AI-generated outputs requires careful scrutiny because of the lack of clarity around copyright ownership of AI-generated works. Did the vendor’s humans contribute to the output or was it purely AI-generated on their end?
In Naruto vs. Slater (2011), referred to as “the monkey case” throughout the webinar, the Ninth Circuit Court of Appeals ruled that animals cannot be copyright owners under U.S. law, establishing that human authorship is a fundamental requirement for copyright protection.
In Thaler v. Perlmutter (2025), the U.S. Court of Appeals for the District of Columbia Circuit affirmed that works generated solely by artificial intelligence without human creative input cannot receive copyright protection under U.S. law. But the courts have left open the question of how much human involvement is necessary to make AI-generated work eligible for copyright protection.
Some vendors claim to own or assign rights to outputs, but this assignment often has limited practical value. The computer generates the output without human intervention, so there may be no copyright to assign in the first place.
Plus, how much of the output was created from the input? Laura J. Greenberg, Head of AI Legal Academy for Wordsmith explains, “the more specific you are [with your input], the better the results that you’re going to get. So that’s the first key tenant of using AI is that the more specific you are the better your results that you’re going to get.”
💡Pro tip: When negotiating AI terms about output ownership, seek clarity on what the vendor is actually assigning. If the terms contain language allowing similar outputs for other customers, consider whether confidentiality protections might be more valuable than ownership.
4. Data Use Restrictions for AI Training
Will your vendor use your data to train their AI system? This critical question demands clear contractual restrictions.
“I don’t like uncertainty in contracts. I think if you’re going to train, say it. Do not leave that door open,” advised Tollen.
One webinar attendee shared their preferred language from the customer side as follows:
Vendor agrees that it will not train its AI features using Customer Data and has and will continue to have contractual agreements in place with its AI partners which prohibit AI partners from utilizing Customer Data to train or improve their AI models, features, and services. AI partners are required to ensure Customer Data is encrypted in transit and at rest and delete the Customer Data after processing.
Several attorneys joined the conversation and agreed that this clause would be effective in clarifying the vendor’s ability to use the customer’s data to train its AI.
💡Pro tip: When reviewing these provisions in AI contracts, look for comprehensive definitions and explicit limitations. Vague language allows vendors to potentially use your data in ways you didn’t intend.
Connect and Learn: One webinar attendee, Nate Kostelnik, summarized their three key learnings from the discussion.
5. Confidentiality and Usage Restrictions
Effective AI agreements should treat both inputs and outputs as confidential information. However, confidentiality alone doesn’t provide complete protection in AI vendor relationships. In addition to confidentiality protections, the contract should also include usage restrictions about what the other party cannot do with the data.
Tollen emphasized, “The strongest thing is to say they’re not going to use it for any purpose other than to provide you with the service. That goes beyond confidentiality, that goes beyond IP.”
For customers, you should specifically state that the vendor cannot train their AI using your customer data. Remember that ownership, confidentiality, and use restrictions are separate concepts requiring distinct contract provisions.
By understanding these key issues from both customer and vendor perspectives, legal professionals can negotiate more balanced and effective AI vendor agreements while leveraging AI tools to enhance their own contract review processes.
Continued Learning Opportunities
- You can now access the full 75-minute webinar recording, which comes with the presentation deck containing example screenshots and redlines, as well as bonus materials.
- For more expert-led webinar recordings about contracts, please visit the Contract Nerds Video Library.
- Join 20,000+ lawyers and contracts professionals who want to master contracting skills by subscribing to our weekly newsletter.
