The Copyright Office has settled the easy half the question. But the other half — the half where most event marketers live — is still up in the air.

Who owns the rights to an image AI generates — a conference logo, say, or an event theme graphic?

According to Barbara Dunn of Barbara Dunn Law, speaking on a recent SITE Florida & Caribbean webinar hosted by HopSkip, if the output is 100% AI-generated, it likely isn’t protectable at all. The Copyright Office already has made it clear that 100%-AI-generated outputs are not subject to and can’t benefit from the protections of copyright.

The test case involved a computer scientist, Stephen Thaler, who applied to register an artwork he said his AI system generated autonomously, naming the machine as author. The Copyright Office refused. A federal district court upheld the refusal in 2023, holding that human authorship is essential to a valid copyright claim. In March 2025, the D.C. Circuit affirmed, reading the Copyright Act’s provisions on ownership, duration, inheritance and signature requirements as presupposing a human being. The full court declined to rehear the case that May, and the Supreme Court denied review.

So it is settled that an image, logo or theme graphic produced entirely by an AI system, with no meaningful human creative contribution, cannot be registered — and a work you can’t register is a work you will struggle to enforce against anyone who copies it.

What’s still a bit murky is when AI generates less than 100% of the output. For example, she said, what if 70% of an article was written by a staff member and 30% was massaged by AI — the most likely scenario for most event marketing outputs right now. Is the result protectable?

The Key May Be Creative Control

The answer may lie in who, or what, maintains creative control, according to the Office’s registration guidance, first issued in March 2023 and expanded in the copyrightability section of its Copyright and Artificial Intelligence report in January 2025. The human, not AI, must be the one who determines the expressive elements of the output — and no, prompting doesn’t count, at least not enough to push the output into a copyrightable category.

A couple of registration decisions that may provide some illumination. In the case of graphic novel Zarya of the Dawn, which has AI-generated illustrations, the Office found that the individual images were unprotectable. However, it did allow the human-written text and the overall selection and arrangement of images and text to be copyrightable. That novel in some ways is like a conference program, an event website or a printed guide: The human-authored copy and the human curation of the whole may be protectable even where individual AI-made images are not.

Another relevant case is when an applicant tried to copyright an original photo he had taken, and then used an AI to enhance the image. Because the AI decided how to combine the filter and the image, the Office nixed the copyright. While using a closed, enterprise AI system that uses the organization’s own past logos, colors and styles may support the argument that the result wasn’t 100% AI-generated, this second case cautions that it may not be enough to just provide the source materials and style to reach the creative control threshold for copyright. Another decision, this one involving the AI-assisted artwork Théâtre D’opéra Spatial, failed because the applicant didn’t identify and disclaim the AI-generated portions, as the guidance requires. That case is now in litigation. Groups planning to register anything AI-assisted should know that disclosure is a condition of registration, not an optional courtesy — which in turn means someone has to be keeping records of how AI was used. After all, the Office has registered hundreds of works containing AI-generated material since the 2023 guidance — with protection covering the human contribution. Using AI doesn’t forfeit copyright in what a person actually authored. It narrows what the registration covers.

Unprotectable Versus Infringing

Unprotectable is one thing. Infringement is another thing entirely, and it’s important to know the difference. If an output is unprotectable, the group can’t stop anyone else from using that output. That’s why groups hire a graphic artist precisely — so it can own the branding around its event. Dunn advised that whoever creates it should assign their rights to the group in writing. An AI system doesn’t sign an assignment, and a fully AI-generated logo may leave the group with nothing enforceable to assign.

Infringement is when someone else brings a claim against the group. Dunn said she has had multiple clients pulled into disputes over photographs — an image of a destination or venue that seemed fine to use, until the owner surfaced with a demand. AI can further complicate this scenario.

The legal test for an infringing output is that the AI had access to the underlying work, which may be shown by evidence it was trained on that work, and that the output is substantially similar to it. Both the user who prompted the output and the AI company can face liability. Two categories carry elevated risk: outputs involving existing fictional characters, which can enjoy protection independent of the works they appear in, and outputs generated in the style of a named artist — though style itself is generally not protected, and some image tools now decline those prompts outright.

Dunn’s test for anything the organization puts into market remains the practical one: do you own it outright, or do you have a license — which, as she puts it, is the fancy word for a permission slip? If neither, the group is in a weak position.

How Hotels Are Handling It

During the webinar, Kelly Bagnell of Holland & Knight, who represents hotel management and ownership, said that hotels face third-party claims over image rights regularly, and those suits are painful because, by the time a claim arrives, the material is already deployed, and undoing it costs a fortune. That drives an aggressive approach to ownership — confirming rights to every asset going into market under the brand.

Her read on AI-generated material, offered as an educated guess rather than a description of client practice, is that it wouldn’t be used formally at all. The standard is ownership of every single element, and AI output doesn’t clear it.

Don’t Assume the Vendor’s Indemnity Covers You

Where planners have direct leverage is in agreements with vendors on the event food chain — event tech, registration, marketing and design partners. Dunn’s checklist for an AI clause covers ownership specifically:

• Define AI broadly. And if the vendor operates a closed system, say so in the definition.
• State who owns the output. Don’t leave IP ownership of work product to be sorted out later.
• Require disclosure of AI use. Set out what the vendor may use AI for and what it may not.
• Require records. Needed both to reconstruct a problem and to support any registration application.
• Name the human oversight. Where a person reviews the output before it goes anywhere.

Bagnell added a caution that applies to any clause of this kind: It is only as good as the organization’s internal communication about it. If everyone involved doesn’t know the parameters, even the most well-crafted provision could fail.

While the practical takeaway is that it is pretty clearly not acceptable for an organization to ask AI to generate the event program — or the logo, or the theme art, or the website layout — 100%. What’s still being hashed out now is how much human involvement is enough, and the registration decisions to date suggest the threshold is higher than most people assume. Groups that document who did what, and when, will be in a materially better position than groups that don’t — whether they end up filing a registration or defending a claim.