Your Team Delivered Work Made Partly With AI. What Can You Promise the Customer About Who Owns It?

Your agency, studio or development shop delivered the work: a codebase, a campaign, a set of designs, a report. Some of it was drafted, generated or reshaped with AI tools along the way. Now the customer's contract asks you to assign everything you made, warrant that it is original, and defend them if anyone claims otherwise.

Those three promises were written for a world in which people made all of the work. This post is for the company on the delivering side. It explains what each promise can do when AI was involved, where the gap opens between what the clause says and what you can deliver, and what to settle before you sign. It is a general discussion of contract practice and related U.S. copyright principles, not advice about what any particular agreement should contain.

Three promises, three different jobs

Services agreements often address ownership and related IP risk through three distinct provisions:

1. An assignment. You transfer to the customer whatever rights you hold in the deliverables and agree to assign.
2. An originality warranty. You promise that the deliverables meet the agreement's stated requirements for originality or noninfringement.
3. An indemnity. You agree to cover specified third-party infringement claims, potentially including defense obligations and covered costs.

Each clause does a different job, and AI affects each one differently. Treating them as one "IP clause" is where providers get into trouble.

What an assignment can carry

An assignment moves rights from you to the customer. It cannot move rights that do not exist. Our post on who owns AI-generated code and content covers the human-authorship rule. The short version is that AI-generated material may lack copyright protection, while human-authored expression, creative selection or arrangement, and sufficiently original human modifications may be protected. The scope of protection depends on the human contribution to the final work, not simply on whether the team calls its process "AI-assisted."

For a provider, that raises three practical questions:

• Which parts of the deliverable did your people create, select, arrange or revise, and which parts came out of a tool unchanged? That helps identify potentially protected human contributions. What the assignment carries also depends on which rights you hold and what the clause transfers.
• What do your own tool licenses say about output? Some AI providers' terms address who may use output and on what conditions; our post on customer contract terms for AI products covers how those terms are commonly structured on the product side.
• Does the clause promise more than a transfer? A clause that says the customer "will own all right, title and interest" may promise an ownership outcome rather than itself effect an assignment. Check both the transfer language and any separate ownership assurance. A clause that says the deliverables "will be protectable by copyright" or "registrable" is an additional promise, and one a provider may not be able to keep for AI-generated portions.

None of this means AI-assisted work is worthless to the customer. It means the assignment should describe what it transfers in terms that match how the work was made.

What "original" can mean

An originality warranty is where the AI question is sharpest, because customers may attach different expectations to the word. The clause should distinguish among these expectations rather than assume they are interchangeable:

• Not copied. The customer may expect assurance that the deliverables do not infringe third-party rights. An AI tool can produce text or code resembling existing material without the provider recognizing the source. Resemblance alone does not establish infringement, and permission is not the only basis on which third-party material may lawfully be used.
• Made by us. The deliverables are the provider's own work rather than a third party's. If a tool generated a portion, is that portion "made by us"?
• Protectable. The deliverables qualify for copyright. As above, that is a legal conclusion the provider may not be in a position to give for generated material.

Before you give the warranty, decide which meaning you are giving, and consider saying so. A provider that knows it used AI can describe its process honestly: what was generated, what was reviewed, what was checked against third-party sources. Our post on a contractor's use of AI sets out the checks a customer may ask about, including the review of code that may carry an open-source license. The provider's job is to be able to answer them.

What an indemnity commits you to

An indemnity allocates specified risks and may include a duty to defend, reimburse covered losses, or both. For AI-assisted work, start with three features that affect your exposure:

• Scope. Does it cover any claim that the deliverables infringe, or only claims arising from your breach of the originality warranty? A warranty-based trigger may narrow the obligation, but only if the warranty and indemnity are themselves limited. It does not necessarily confine exposure to matters you controlled.
• What you receive upstream. Some AI providers promise to defend their customers against certain claims about output, on conditions. Our post on what an AI provider's promise to defend you asks of you explains those conditions. If your indemnity to the customer is broader than the protection you receive upstream, you may assume obligations your AI provider's indemnity does not cover. Compare the triggers, exclusions, conditions, defense procedures and limits—not just whether both contracts contain an indemnity.
• The cap. Indemnities are often carved out of the liability cap. Our guide to how much your liability cap should be explains how caps and carve-outs interact, and why an uncapped indemnity for AI-assisted work deserves a second look.

Telling the customer

Whether to tell a customer that AI was used is partly a contract question and partly a relationship question. On the contract side, check three things:

• Does the agreement already address it? Some customer templates now prohibit AI use in deliverables, require disclosure, or require that specific tools not be used with the customer's material. Our post on why employee NDAs and ChatGPT do not always mix covers the confidentiality side of putting a customer's material into a tool.
• Does the customer have its own policy? Enterprise customers increasingly have AI-use policies that reach their vendors. Our post on AI subprocessors and customer contracts covers how those commitments flow down.
• What do your own people know? A disclosure promise is only as good as your internal record of who used what. Our post on a workplace AI use policy covers the internal side.

Keep a record

Most of the promises above depend on being able to show how the work was made. Drafts, commit history, prompt logs where your tools keep them, and notes of what was checked against third-party sources all help a provider answer an originality question or support an assignment later. Keep those records subject to confidentiality, privacy, security and retention requirements; do not retain sensitive prompts simply for completeness. Records can support the analysis, but do not themselves create copyright or establish ownership. Our post on AI in investor diligence explains why the same records matter when a customer, or your own investor, asks.

Subcontractors

If part of the work came from a subcontractor or independent contractor, check both the rights you obtained and the promises they made to you. Your obligations to the customer can exceed what the subcontractor owes you. You cannot transfer rights you neither own nor have authority to transfer, and your warranty may cover subcontractor work even if you did not supervise how it was made. Our post on the difference between an MSA and a SOW covers how the master terms and the project terms fit together, and our guide to the contracts every startup needs covers the contractor agreement itself.

Questions to settle before you sign

These are questions to take to your contracts, not conclusions about any particular agreement:

1. Which parts of the deliverables did our people create, select or revise, and which came out of a tool unchanged?
2. Does the assignment transfer rights, or does it also promise that the rights exist?
3. Which meaning of "original" does the warranty use, and can we give it for every part of the work?
4. What does our indemnity cover, and how does it compare with what we receive from our AI providers?
5. Is the indemnity inside or outside the liability cap?
6. Does the customer's agreement or policy restrict or require disclosure of AI use, and does our process match it?
7. What do our subcontractors promise us, and does it support what we promise the customer?
8. Could we show how the work was made if asked a year from now?

What this post does not cover

This post does not say whether any AI-assisted deliverable is protected by copyright, whether any output infringes, or how any clause would be enforced. It does not cover trademarks or patents, employee invention assignments, or the customer's side of the negotiation, which our contractor post addresses. Our post on AI legal risks for startups covers the wider set of questions.

How we can help

Accord & Shield Legal drafts and reviews services agreements, customer contract terms and intellectual property assignments for technology companies in Arizona, California and Texas. To better understand the scope of your matter, you can schedule an initial consultation. The consultation is evaluation-only; no legal advice is provided. Representation begins only after an engagement agreement is signed and any required retainer is received. Please do not send confidential or sensitive documents before we confirm that we can assist.

---

This article is general information, not legal advice. Reading it does not create an attorney-client relationship. Whether a particular deliverable is protected, and what a particular clause promises, depends on the work and the agreement; consult an attorney about your situation.