Your Agency Used AI. Who Owns the Output?

Your agency's contract almost certainly says every deliverable becomes yours on final payment. That clause still works — for everything that has an owner to begin with. Where a tool produced the material outright, there may be no copyright in it for anyone to assign, and the strongest sentence in your agreement quietly transfers nothing. This is not an argument against agencies using AI. It is an argument for knowing which of the things you paid for are protected by your contract and which are protected by something else entirely.

What does "all IP transfers to you" actually transfer?

An assignment clause moves rights that exist. It cannot create them. In January 2025 the US Copyright Office published the second part of its report on copyright and artificial intelligence and concluded that the outputs of generative AI can be protected "only where a human author has determined sufficient expressive elements" — and that this does not include "the mere provision of prompts" (US Copyright Office, January 29, 2025). Now read that against your contract. If a designer typed a description of your logo into a tool and sent back what came out, the clause assigning you the copyright is assigning a copyright that was never created. You own the file. You may own nothing about the design.

What did 2026 settle?

The question stayed open for a while because it was still moving through the courts, and plenty of agencies used that to wave it off. It is now closed at the top. On March 2, 2026 the US Supreme Court declined to hear Thaler v. Perlmutter, leaving standing the rule that the Copyright Act "requires all eligible work to be authored in the first instance by a human being" (Morgan Lewis, March 6, 2026). No new legislation replaced it, and the Copyright Office's January 2025 position is still the one applications are examined against. For your next planning cycle, treat human authorship as a settled requirement rather than a pending question.

Which deliverables are actually exposed?

Not all of them, and not equally. Sort what you commissioned into three piles. The first transfers regardless of how it was made: your domain, your repository, your hosting and analytics accounts, your customer data, your trademark registrations. None of that depends on copyright at all. The second is human-authored work that used AI somewhere along the way — a designer who generated forty directions, chose one and rebuilt it; a developer whose editor suggested lines they read, rejected and changed. The Copyright Office is explicit that the use of AI to assist in the process of creation "does not bar copyrightability" (January 29, 2025). That pile is fine. The third is material that came out of a tool substantially as delivered, and it is the only exposed one. The person who knows how big that third pile is on your project is your agency, and they will not raise it unprompted.

What should you ask for before final payment?

You do not need a bespoke legal instrument for this. You need your agency to tell you what it did and to put its name on the answer, which costs an honest studio nothing. The list below is what we would want a client to ask us, and most of it is what we volunteer before anyone asks.

Should you ban AI from the project?

No, and a blanket ban would cost you more than it protects. The same guidance that withholds protection from prompt-only output preserves it for work where a person made the expressive decisions, which describes most of how competent studios actually use these tools. A ban raises your invoice and changes nothing about ownership, because what matters is whether a human authored the expression, not whether software was open at the time. One caveat, and it is a real one: NexusWave is a studio in Vancouver, not a law firm, every source above is United States law, and Canadian, Iranian and Gulf practice all differ. None of this is legal advice. What we can give you is a written account of what we did on your project — which is the part most agencies are not volunteering.