Legal and Copyright Considerations for AI Video in Commercial Advertising

AI video in commercial advertising involves two separate legal questions. Can the brand own the work, and does it have permission to use it commercially? A platform subscription does not answer both. To protect its work and use it legally, a brand needs human authorship, license terms that cover its use, clearance of third-party rights, and a review of every asset before it reaches paid media.

Why the March 2026 Supreme Court ruling changed the ground rules for advertisers

The U.S. Supreme Court chose not to hear Thaler v. Perlmutter in March 2026, leaving the D.C. Circuit’s ruling intact. The denial largely settled the narrow question behind the case, rather than creating a new rule through a decision on the merits. AI cannot hold copyright, and the work must have genuine human authorship to be protected. For advertisers, that is the starting point. Paying for a generated video does not make the buyer its author.

The U.S. Copyright Office reserves protection for work in which a human made the meaningful creative choices. Its rejection of prompt-only authorship claims dates back to the Zarya of the Dawn decision, so the outcome was not a sudden reversal. The distinction is between requesting an output and authoring it. When the model decides how to compose the work, the person who gives it instructions does not automatically get copyright in the result.

The EU reaches a similar result through a different standard. The CJEU’s “free and creative choices” test does not treat model-made compositional decisions in response to a prompt as sufficient originality. The UK is a notable exception. Its Copyright, Designs and Patents Act allows copyright in computer-generated works, so brands running campaigns there have different rights from those under the U.S. rules discussed here.

What it means in practice that a purely AI-generated ad video is effectively in the public domain

A product video made entirely from a text prompt, without human scripting, editing, or arrangement, has no human-authored work for copyright to protect. The brand may have paid for it. It may also have spent considerable time getting the output it wanted.

The commercial consequence is blunt. A competitor can take that purely generated video and use it without the original brand having a copyright claim to stop the copying. Other rights, including likeness and trademark rights, remain separate questions, but copyright supplies no exclusivity in the AI output itself. For a brand investing in a distinctive visual identity, that is a structural weakness.

The reverse is also true for other brands’ purely generated material. A business can use it without copyright concerns over the AI-generated elements, though that freedom rarely makes up for losing control of its own creative assets. More prompting does not fix the problem. The Copyright Office’s 2025 guidance says that with the technology now available, prompts alone do not give a human enough control to establish authorship. Detail and repeated revisions to a prompt are not substitutes for creative control over the finished work.

How human creative contribution changes what a brand can actually own

Human creative work changes the ownership position, but only for the parts it actually shapes. The Copyright Office protects the human-authored parts of AI-assisted work but leaves the AI’s outputs unprotected. That distinction makes production choices legally significant.

In a video workflow, human contribution can include an original script, scene direction, footage editing, pacing, color grading, and final composition. Original audio, brand-specific overlays, and hand-crafted typography can add further human-authored elements. Selecting among generated variations based on documented aesthetic judgment can also form part of that creative process. The photographer analogy helps explain the principle. Using a machine does not make the machine the author when a human meaningfully controls the creative result.

That control needs a record behind it. Logs of prompts, rejected variations, editing decisions, and original contributions help establish what the human actually authored if ownership is challenged. Prompt records show how the work was made, but they do not prove that prompting alone creates copyright. Voyager’s persistent workflow memory, which learns how users like work done, supports this documentation within a multi-model creative workflow.

For brands seeking ownership, the lesson is to design human contribution into production and document it. One-click generation offers speed, but it does not supply the same basis for a copyright claim.

Why copyright ownership and commercial use rights are two entirely separate legal instruments

Copyright determines who can stop others from copying protected work. A commercial license gives contractual permission from a platform to use its output for revenue-generating purposes. These rights operate independently of each other. A brand can have permission to run a video in paid advertising without owning copyright in that video, and human authorship does not excuse a breach of platform terms.

The practical instruction is to check the actual commercial grant before publishing. Free and trial plans often do not allow commercial use, and some platforms have different rules for paid advertising and organic social content. The license may need to cover client delivery and resale separately. The underlying stock footage, audio, and imagery also need a clear commercial rights chain. Permission to generate something is not enough to establish permission for every intended use.

Even a suitable license has limits. It governs the brand’s relationship with the platform, while claims a third party might bring once the campaign goes public may fall outside its scope. Enterprise procurement increasingly seeks copyright indemnification, under which a platform covers legal costs and damages for covered infringement claims involving generated output. Vendors offer this protection inconsistently, and it often needs negotiation. It provides a contractual backstop alongside review of the work.

Multi-model production adds another contractual complication. Each model can have its own terms, training-data provenance, and commercial-rights structure. A unified license that covers multiple models can simplify that work, but the scope of the grant still determines what the brand is allowed to do.

Training data liability and why it falls differently on platform users versus internal model operators

Training-data litigation is primarily a platform-level risk when a business uses a reputable vendor that claims lawful training data. That distinguishes the position of a platform customer from that of a brand taking direct responsibility for a model, while the safety of individual outputs remains uncertain. Brands need to establish where responsibility sits in their production setup.

That position changes when brands train or operate models internally, including open-source models with unclear training-data sources. Improperly licensed material can increase the brand’s exposure because it is no longer relying only on a vendor-operated service. The rights chain becomes more consequential. Coverr’s stock-to-AI workflow illustrates a different approach, anchoring generation to a licensed, commercially cleared source library. That gives brands a written record of the source-rights chain to check.

For legal and procurement teams, the instruction is to understand how the vendor’s model was trained and build appropriate indemnities and clearance provisions into the master services agreement. The finished clip still needs scrutiny. “The AI made it” is not a defense against infringement claims. If an output closely resembles a specific film, advertisement, or artist’s identifiable work, the exposure does not disappear because a third-party tool produced it.

Likeness rights, voice cloning, and the consent requirement that platform licenses do not cover

A commercial platform license does not give a brand permission to use someone’s identity. The right of publicity governs commercial use of a person’s name, image, likeness, and voice separately from copyright. Synthetic production does not erase that distinction.

State law sets most of the rules for U.S. publicity rights. Tennessee’s ELVIS Act was the first to cover AI voice clones, while California and New York have strong protections. California amended its law in 2024, effective January 1, 2025, to cover AI replicas of deceased performers. The proposed federal NO FAKES Act would create a national right against unauthorized digital replicas, but it has not passed.

The practical rule is to obtain written consent for any real or recognizable face or voice used in commercial advertising. Public figures are not an exception. A synthetic presenter made to look like an identifiable person needs the same consent.

Trademark rights create another gap that the platform license does not close. Generated video can reference or imitate protected names, logos, and trade dress, particularly in retail advertising. Those elements need their own clearance before the ad runs. Neither a commercial-use grant nor a copyright claim in human edits resolves a third party’s rights in its brand identity.

Disclosure obligations that apply to AI-generated video in advertising, and how they interact

Each video must satisfy the disclosure rules that apply to it. Every version produced after campaign-level approval therefore needs disclosure review, especially when generated performers or personas appear in more than one version.

New York’s Synthetic Performer Disclosure Law comes into effect on June 9, 2026. It requires producers of advertising content to conspicuously disclose synthetic performers, meaning non-identifiable human likenesses generated by AI or algorithms. Identifiable replicas raise consent questions, while synthetic performers raise this distinct disclosure question. Because of digital ad targeting, the law reaches most national U.S. campaigns in practice, including those that are not aimed expressly at New York audiences.

Article 50 of the EU AI Act adds transparency duties for AI-generated content, and its rules are being enforced in phases. Those duties include visible labeling, machine-readable marking, and transparency declarations where applicable. The technical format needs careful treatment. The Act does not name C2PA in its binding text, although C2PA is a leading example in supporting guidance. Supporting a particular format is therefore not, by itself, a complete answer to compliance.

For production teams, the resulting review reaches into model selection and the pipeline. It needs to check whether the platform supports the provenance marking needed, where disclosure metadata is added during production, and who agrees in the contract to take responsibility for non-compliance. Each dynamically generated variation containing AI personas or testimonials is a separate disclosure event. Healthcare, financial services, legal, and real estate advertising must also satisfy their sector-specific rules, so AI disclosure does not replace the obligations already attached to those ads.

How workflow design — not just tool selection — determines a brand’s actual legal exposure

The same tool can produce different legal risk profiles depending on how a team uses it. A prompt-only output and a video with documented human scripting, editing, and composition do not present the same ownership claim. A suitable commercial plan and an unsuitable trial plan do not provide the same contractual permission. The review process has to capture those differences before an asset reaches paid media.

Creative directors and legal teams need a review they can repeat before work is published. It should establish the documented human contribution and the provenance of the model’s training data, then match the commercial license to paid advertising, client delivery, and the intended channels. Consent and unintended resemblance need review for faces and voices. Names, logos, and design elements need trademark clearance, while disclosure review must cover FTC, applicable state-law, platform-specific, and EU requirements at the asset level.

That review must also reach beyond the primary tool. When teams chain models for scripting, generation, and post-production, each model’s terms belong in the assessment, together with appropriate indemnities and clearance provisions. Voyager’s unified multi-model dashboard and governed relationship can simplify that contractual work, while persistent workflow memory supports the authorship record.

Enterprise protection also depends on negotiated indemnification, vendor agreement provisions, and internal approval routing. The review categories above give counsel an agenda grounded in the actual asset and its production history. High-budget AI video assets should go through legal counsel before campaign launch.