A brand protection team drafts a cease-and-desist letter over a copycat ad campaign — same layout, same tagline structure, same visual mood. Before it goes out, someone asks a routine question: who made the original? The answer traces back to a marketing contractor who ran a series of prompts through an image generator last year, made light edits, and shipped the result. No revision log. No record of what a human actually changed. The letter still gets sent. Whether the underlying asset was ever eligible for copyright protection in the first place is a question nobody stopped to answer.

That gap — between what gets enforced and what can actually be proven to be owned — is becoming a routine feature of brand protection work, not an edge case.

The Visible Problem: Ownership Is Uncertain Before Infringement Even Enters the Picture

If an AI tool generates an image, a slogan, or a video, the immediate question isn’t who copied it. It’s whether anyone owns it at all.

• The company that ran the prompt?
• The person who wrote it?
• The developer of the underlying model?
• Or no one?

This isn’t an abstract debate. It determines whether a brand has any standing to act when a competitor, counterfeiter, or opportunistic account reuses the same creative territory.

The Deeper Problem: Teams Are Enforcing Against Assets Whose Own Status Is Unresolved

Marketing and legal teams have largely kept moving at the same pace they always have — publish, monitor, enforce — while the underlying production process has shifted underneath them. AI-assisted drafts, mockups, and campaign variants now move through creative pipelines with little to no record of where the machine’s contribution ended and a human’s began.

The consequence cuts in two directions. First, brand teams may be building enforcement cases — cease-and-desist letters, DMCA notices, opposition filings — on assets whose copyright status was never resolved internally. If challenged, a brand may not be able to show the “meaningful human authorship” that U.S. law requires. Second, the same uncertainty works against enforcement: when a counterfeiter’s output is itself AI-generated, doctrine hasn’t caught up enough to say clearly whether that output infringes, or whether it’s simply unprotectable in a different direction. Neither side of the equation is settled, and most brands haven’t documented enough of their own process to know which side they’re standing on.

1. Only Human Authorship Is Copyrightable

The U.S. Copyright Office has been direct on this point: only works with human authorship qualify for copyright protection. A work generated entirely by AI, published as-is, is not eligible for registration — regardless of how much creative direction went into the prompt.

2. Partial AI-Generated Works Can Still Be Protected — If the Human Contribution Is Real

Where a human contributes something creative, original, and substantial enough to make them a co-author or dominant contributor, the resulting work can qualify. The Copyright Office evaluates this case by case, and the line is narrower than most teams assume.

Likely to qualify:

• An AI-generated image substantially edited or combined with human-created elements
• A script drafted by AI, then meaningfully revised and rewritten by a person
• AI output used as raw material inside a human-directed creative process

Unlikely to qualify:

• An AI-generated image, article, or track published directly from the output
• Cropping, brightness adjustments, or a new title layered on otherwise untouched AI work

If a brand cannot point to which parts of an asset were human-made, it may not be able to claim copyright in it at all — which means it may not be able to stop anyone else from using it either.

Case in Point: “AI Barbie” and the Cost of Moving Fast on Someone Else’s IP

After the Barbie film’s release, the “AI Barbie” trend spread across social platforms, with users running their own photos through Midjourney and similar tools to generate stylized, boxed-doll versions of themselves — complete with captions and Barbie-branded packaging cues. The appeal was obvious. The exposure was less obvious to the people posting.

Barbie’s name, likeness, packaging, and signature Pantone 219C pink are protected trademarks and copyrighted material, and Mattel has treated the trend as an enforcement matter, not a compliment, reportedly pursuing claims tied to the character’s protected elements.

Why It Matters

Mattel has issued DMCA takedown notices against creators, developers, and apps that recreate Barbie’s likeness without permission — a reminder that a tool’s ease of use has no bearing on whose IP it’s built on. A post captioned “this Barbie is a CEO” reads as a joke; to a rights holder actively enforcing, it reads as unauthorized use of a protected character.

AI Is Already Inside the Examination Process, Not Just the Content

Brands aren’t the only ones running AI through IP workflows. Trademark and patent offices are using it to review the very filings brands submit.

The USPTO uses tools like “Similarity Search” and “More Like This” to help examiners surface visually or conceptually similar marks faster. Since September 2022, examiners have run over 1.5 million AI-assisted searches. That scale changes the practical odds that a mark resembling existing IP gets flagged early, whether or not the applicant’s own creative process involved AI at all — one more reason a brand’s internal authorship record matters before a filing, not after a dispute.

The Harder Question: Can You Enforce a Right You Can’t Prove You Hold?

Here is the version of the problem most brand and legal teams haven’t sat with directly. If a company’s own marketing asset was AI-assisted without a documented human contribution, and a competitor or infringer copies it, what exactly is the company enforcing? A cease-and-desist letter assumes an underlying right. If that right was never secured because the asset wasn’t copyrightable in the first place, the letter is a bluff dressed as a legal position — one that unravels quickly if the other side pushes back and asks for proof of authorship.

This isn’t a hypothetical for the future. It’s a live question for any brand that has run creative through AI tools over the past two years without tracking who did what.

What Better Looks Like

Until case law and statutes catch up, the practical fix sits inside the workflow, not the courtroom:

• Document the AI workflow as it happens. Record which tools were used, what was generated, and what a person changed — not after a dispute arises, but as part of production.
• Make the human contribution real and traceable. Editing, combining, and transforming AI output strengthens a copyright claim; light touch-ups don’t.
• Vet tools before adopting them. Understand training data provenance and usage rights, since some AI platforms carry their own IP exposure.
• Loop in counsel before publication, not after a competitor copies the asset — particularly for anything that will later need to be enforced.

Questions Worth Asking Before the Next Takedown Notice

• Can we show, asset by asset, which parts of our current marketing library were human-made versus AI-generated?
• If a competitor copied our last AI-assisted campaign tomorrow, could we prove we own what they took?
• Do our vendors and contractors document their AI use, or does that record exist only in their heads?

Answering these before an infringement dispute is far cheaper than answering them during one.

Investigating who else is using a brand’s assets, and how closely, still depends on being able to see the pattern across platforms in the first place. That’s the piece of this that tooling like Hubstream is built for: an AI-native environment for tracing where infringing content surfaces, comparing it against a brand’s own material, and prioritizing which cases are worth acting on — which only matters once a brand can say with confidence what it’s comparing from a position of clear ownership.

The team drafting that cease-and-desist letter will send it either way. The open question is whether, the next time someone asks them to prove what they own, they’ll have an answer ready before the letter goes out — or after someone challenges it.