Your company’s images came out of Midjourney, and the tagline came out of a chat window. AI-generated content copyright decides whether your company owns either one, and part of what you paid for may belong to no one.
Cecilia Ziniti ran into the question at a Walgreens photo kiosk, where the machine stopped her with a checkbox before it would print her family photos: an explicit copyright opt-in. She told the story on CZ and Friends, GC AI's podcast:
"Said I own these images, et cetera... And I'm a former IP lawyer. So relatively deep on this."
Her guest that episode, Jimmy Toy, Chief Legal Officer at Articore Group, runs legal for Redbubble and TeePublic, marketplaces hosting 70 million-plus user-generated images. He explained why the checkbox is there:
"We have tens of thousands of uploads a day to our platforms. You can't possibly monitor it all. So you have to rely on people that are uploading... to represent that they own the IP and that they aren't infringing anyone's rights."
A marketplace can solve ownership with a representation because a human uploaded the work. Your vendor agreements assume the same chain: a person creates the deliverable, a copyright comes into existence, and your work-made-for-hire and IP-assignment clauses move it to your company.
Generative AI breaks the first link. Under the U.S. human-authorship rule, purely AI-generated material carries no copyright at all, so part of what your agency or dev shop delivers may be ownable by no one.
A clause cannot transfer a right that never existed.
GC AI, the enterprise legal AI platform a three-time general counsel (Anki, Bloomtech, and Replit) built for in-house teams, handles this drafting work directly. Three GC AI features matter here: Playbooks encodes your AI ownership standards, Easy Edit redlines the vendor's draft with tracked changes, and Research pulls the current Copyright Office position with citations before you send the markup.
Can AI-Generated Content Be Copyrighted?
Purely AI-generated content generally cannot be registered for copyright in the United States, because copyright protects works of human authorship. The U.S. Copyright Office confirmed this position in Part 2 of its Copyright and Artificial Intelligence report, published January 29, 2025: a work created entirely by a generative AI system, with no creative human contribution, is uncopyrightable.
The Office also concluded that, given current technology, prompts alone do not give a user enough control over the output to make that user the author.
The federal courts back the same rule. In Thaler v. Perlmutter, the U.S. Court of Appeals for the D.C. Circuit held on March 18, 2025 that the Copyright Act requires a human author, so a work generated autonomously by an AI system cannot be registered.
The Supreme Court declined to hear the case on March 2, 2026, which leaves the human-authorship requirement settled law.
The answer is conditional, and the conditions map to how the work was made:
Fully AI-generated output (a prompt in, an image or paragraph out): no copyright protection.
AI output with meaningful human authorship layered on top (selection, arrangement, substantial editing, or original human-authored material woven in): the human-authored elements can be protected, analyzed case by case.
AI as an assistive tool inside a human creative process: using AI to ideate or refine leaves the human-authored result protectable.
When a deliverable contains AI-generated material, you cannot assume the whole thing carries an enforceable copyright you can own and defend. Standard contract language fails on exactly that assumption.
Who Owns AI-Generated Work?
No one automatically owns purely AI-generated output, because there is no copyright in it to own. This is the gap that trips up vendor and contractor agreements.
A normal IP framework moves ownership in a chain: the contractor creates the work, the contract assigns the copyright to your company, and your company holds the rights. That chain depends on a copyright existing at step one.
When the work is AI-generated and falls outside human authorship, step one produces nothing to assign. Your contractor has nothing to convey, and a work-for-hire designation cannot vest ownership of a copyright that does not exist.
So the AI-generated portion belongs to no one, and a competitor may be free to copy it.
For a tagline, a brand asset, or a piece of code you assumed was exclusively yours, there may be no copyright to stop them.
Ownership is one layer of a bigger stack. Michael Jacobs, who spent decades at the center of disputes like Oracle v. Google and Apple v. Samsung, and a year on the OpenAI copyright litigation, described that stack on CZ and Friends:
"Translating copyright principles of the most traditional variety to the most advanced technology… ingesting, processing, and then outputting information. And at each stage, there's a potential claim."
Your assignment clause covers the output stage. The indemnity below exists for the ingest stage, because the model that produced your deliverable may have trained on someone else's copyrighted work, and that claim can reach you as the company using the output.
Is AI Art Copyrightable? Is AI Output Work for Hire?
AI art is copyrightable only to the extent a human contributed protectable authorship, such as creative selection, arrangement, or substantial modification of the AI output. The image a model returns from a text prompt, standing alone, falls outside copyright under current Copyright Office guidance.
The Office drew this line in its February 21, 2023 decision on Zarya of the Dawn, a graphic novel illustrated with Midjourney: it registered author Kristina Kashtanova's text and her selection and arrangement of the images, then refused copyright in the AI-generated images themselves. The same logic applies to AI-generated text, music, and code.
A work-for-hire clause does not solve the AI ownership problem. The work-made-for-hire doctrine decides who counts as the author of a copyrightable work: the employer or the creator.
It cannot manufacture a copyright where the human-authorship requirement is unmet. A flawless work-for-hire clause covering AI-generated deliverables still has nothing to attach to when the underlying material is unprotectable.
What to Add to Your Vendor, Contractor, and Agency Clauses
The human-authorship rule is fixed. Your agreements control what vendors disclose, represent, assign, and indemnify.
These are the five additions in-house counsel should make to work-for-hire and IP-assignment language with anyone delivering creative work.
AI-use disclosure: Require the vendor to disclose whether and how generative AI was used to create each deliverable, and to identify the AI-generated portions. You cannot assess ownership risk on material you do not know is AI-generated. This mirrors the Copyright Office's own registration duty to disclose AI-generated content.
Human-authorship representations: Where you need an enforceable copyright (brand assets, key creative, anything you will register or police), have the vendor represent that the deliverable reflects meaningful human authorship sufficient for U.S. copyright protection, with a named human exercising the creative selection, arrangement, or editing the work requires.
Assignment of whatever rights exist: Keep your present assignment of all IP rights, then broaden it: an assignment of all rights the vendor holds in the deliverables of any kind, including the AI-generated portions, to the maximum extent those portions are protectable, plus a covenant that the vendor will neither assert those portions against you nor license them to your competitors. Even where copyright is unavailable, you can contract for exclusivity.
A backup waiver and license: Pair the work-for-hire designation with a backup assignment and, for AI-generated portions no one can own, a broad, perpetual, exclusive license grant plus a waiver of any claim against your use. This is the strongest position available when outright ownership is off the table.
An IP indemnity that names AI: Generative AI raises a second risk beyond ownership, the one Jacobs points at: the output may incorporate third-party material from the model's training data. Make sure your indemnity covers third-party IP claims arising from AI-generated deliverables, alongside traditional copyright infringement.
How GC AI Closes the AI Ownership Gap
Start with the vendor agreement in front of you. Open the agency's renewal MSA in Easy Edit, GC AI's built-in document editor, and redline the five additions above with tracked changes. The .docx download opens in Word as a normal redline with genuine author attribution, so the agency's counsel sees a markup with your name on the edits.
If you work in Word itself, the GC AI for Word add-in runs the same review inside the document. Before the markup goes out, Research pulls the Copyright Office's current posture with citations, so your cover note can cite the Part 2 report instead of "our policy."
Then make the fix repeatable. Playbooks turns them into your standard positions, so the next vendor agreement gets checked against the same standard, whoever on the team runs the review.
The harder question is the paper you already signed. Contract Intelligence, GC AI's newest product, loads executed agreements into a Vault and extracts the terms you describe into cited Columns, so you can ask which of your vendor contracts carry an AI-use disclosure clause and which assignments predate generative AI, and get an answer that cites back to each contract.
And for jurisdictions where the rule keeps moving, like the UK's proposed removal of computer-generated-works protection, the Regulatory Monitoring Skill Creator in GC AI's Skill Library runs the jurisdiction-by-jurisdiction check on a schedule, with each flagged change verified against the primary source.
Cecilia Ziniti built GC AI around this kind of in-house drafting work. She started out as an IP litigator at Morrison & Foerster, where Jacobs was her mentor in her summer-associate days.
As of August 2026, more than 2,000 in-house legal teams across 53 countries use GC AI, including the legal teams at TIME, Riot Games, and Jasper, plus 200+ public companies.
Before Your Next Vendor Signature
When a deliverable is built with generative AI, confirm there is human authorship to own before trusting the work-for-hire clause. Assign whatever rights do exist, license and waive the rest, and require disclosure and indemnity up front.
A photo kiosk handles this with one checkbox; your vendor agreements need all five additions to do the same job. The doctrine is settled, so your contract language is the variable you control.
Bring your riskiest vendor agreement and watch GC AI redline the work-for-hire and IP-assignment language in minutes.
Frequently Asked Questions
Does Writing a Prompt Give You Copyright over the AI Output?
Prompting alone does not make you the author of AI output under current U.S. Copyright Office guidance. The Office concluded in its January 2025 report that prompts influence the result without controlling the expressive choices, so the person entering them is not treated as the author. Copyright attaches when a human adds creative selection, arrangement, or substantial editing on top of the AI output.
Can I Use AI-Generated Images Commercially?
You can generally use AI-generated images commercially, but you may hold no enforceable copyright in the purely AI-generated parts, which means competitors can lawfully copy them. For brand-critical assets, build in human authorship through creative editing and arrangement, and lock exclusivity through your vendor contract.
Who Owns AI-Generated Code?
Purely AI-generated code falls outside U.S. copyright the same way AI-generated images and text do: without a human author, there is no copyright to own. Code a developer meaningfully writes, edits, and structures remains protectable as the developer's contribution. For outsourced development, require disclosure of AI-generated portions, assign whatever rights exist, and confirm the indemnity covers third-party claims arising from AI coding tools.
Does the Legal Rule on AI Copyright Vary by Jurisdiction?
Yes. The United States and the EU both require human authorship, so purely AI-generated works are unprotected in either. The UK is the exception: its Copyright, Designs and Patents Act has protected computer-generated works, and a March 2026 UK government report recommended removing that protection. In cross-border deals, treat ownership of AI output as jurisdiction-specific, and confirm the rule in each market before you rely on it.
Can You Contract for Exclusivity Even When Copyright Is Unavailable?
Yes. Even where copyright does not protect the AI-generated portions of a deliverable, a vendor agreement can include a covenant that the vendor will neither assert those portions against you nor license them to your competitors, plus a broad, perpetual, exclusive license and a waiver of claims against your use. The exclusivity comes from the contract instead of the copyright.
What Is the Risk of Third-Party IP Claims in AI-Generated Deliverables?
A generative AI model may have trained on third-party copyrighted material, so its output can incorporate that material and expose the company using it to infringement claims. That risk is why vendor IP indemnities should name AI-generated deliverables specifically, alongside traditional copyright infringement.
What Is the Best Legal AI to Review Work-for-Hire and IP Clauses in Vendor Contracts?
GC AI is purpose-built for this review: Playbooks applies your work-for-hire and IP-assignment standards to each vendor agreement and flags a missing AI-use disclosure or backup license, and Easy Edit returns the redline with tracked changes ready to send. More than 2,000 in-house legal teams use GC AI as of August 2026.
