The short answer is yes, with some nuance worth understanding. In this article we’ll dig into the legal questions clients ask about AI-assisted video. While I am not a lawyer and none of this should be interpreted as legal advice, I have sought the input of Sharon Toerek of Toerek Law, an IP attorney who advises agencies and marketers on exactly these questions.
If you type into your LLM’s prompt box “Make me a funny video of a chipmunk playing the banjo” and receive a vertical clip that you share on your Reels, that is not something you have rights to. The Copyright Office makes a distinction between AI-generated work and AI-assisted work. The former is not copyrightable; the human-created parts of the latter are.
When you hire a studio like IdeaRocket to make an AI-assisted video, we are not just pulling something out of an AI generator. We are doing a lot of human work that is copyrightable. That might include:
- Conceptualization and scripting
- Creative direction
- Substantial image modification
- Substantial editing
Because the human-created parts are protected, someone who filches the video wholesale is still infringing.
However, the elements that did not involve human participation are still not protected. You do not own them. This isn’t something novel; video producers have always included material in their work they did not originate. For instance, a commercial video might include library music or stock footage that is licensed for use, or perhaps material that is in the public domain, or that can be used under fair use doctrine.
There is a narrow strip of instances where the use of AI-generated material might leave you exposed. For instance, if you want to use AI to create a character or mascot that you plan to invest a lot of brand equity in, you should be careful to document the process carefully to prove the salience of human intervention.
Can AI companies use material for training without infringing on copyright?
Apparently, yes. This is not that strange because copyright law has only ever protected the work itself, not its ideas, facts, styles, or techniques.
However, the AI companies have gotten dinged for not paying for access to the source material they used. In Bartz v. Anthropic, the defendants agreed to pay $1.5 billion to settle claims covering more than 480,000 works — the largest copyright settlement in US history. Note, though, that the judge considered the training itself permissible under fair use; the breach was in how the material was obtained.
However, when AIs regurgitate copyrighted content with no modification that is considered a breach of copyright. This is a central claim in the New York Times case against OpenAI and Microsoft, and it was also what GEMA, the music collecting society, claimed in a suit against OpenAI. A German court ruled that not only were AI companies not allowed to reproduce the material, they also couldn’t maintain it in memory, something that isn’t strictly necessary for training purposes.
It’s worthwhile to note that while these cases went badly for the AI companies, the liability did not flow downstream to the users.
On the risk side of the equation, it should also be noted that the legal landscape is fast-changing and ruled by many different jurisdictions, so it’s impossible to say what legal regime your video will one day enter. “The law is always going to be reactive rather than proactive, so it’s always going to be fluid,” says Toerek.
But such risks aren’t the whole picture. “People tend to focus on ownability and infringement of third-party rights,” she says. Toerek points out that other issues also deserve attention.
Sharon Toerek of Toerek Law.
The Need for Transparency
New York State now requires disclosure. Since June 2026, any advertisement running there that includes a synthetic performer has to say so, clearly and conspicuously, with penalties of $1,000 for a first violation and $5,000 for each one after. The obligation falls on whoever produces or creates the ad.
It’s narrower than it sounds. A synthetic performer is a computer-generated figure meant to look like a person who isn’t anyone in particular. Deepfaking a real person is a different law. But the statute never defines “performer,” which leaves an open question about stylized animated characters.
The platforms engaged with this issue first, and for most marketers their guidance is de facto law. YouTube, TikTok, Meta and Instagram all offer a disclosure control at upload. All of them also scan for embedded provenance metadata and will label your video automatically if they find it, whether or not you toggled or ticked anything. Worth knowing: TikTok explicitly exempts artistic styles like anime. On the animated-character question, the platforms are clearer than New York is, and they point a reassuring way.
Beyond distribution, documentation of the production process – which might include a record of prompts used and an itemization of other human inputs – is important if one anticipates a need to protect copyright.
Confidentiality
This isn’t a copyright question at all, which is probably why it gets overlooked. It’s an older one: is your sensitive information being properly safe-guarded? Might it be entered into an LLM that uses it for training purposes?
Both Anthropic and OpenAI draw the same line. Business and enterprise accounts are contractually excluded from model training. Consumer accounts may not be — Anthropic’s consumer plans retain conversations for five years if you allow training, and OpenAI’s train by default unless you switch it off.
At IdeaRocket, we work on business-tier accounts with training disabled, and we require the same of freelancers with exposure to confidential information, rather than assuming it.
Likenesses of Celebrities and Other People
What many practitioners forget is that what was illegal to do before AI is still illegal. You can’t use the likenesses of living people to endorse your product without permission. Says Toerek: “None of the fundamental laws around publicity rights, trademark rights, copyright have changed. There’s just a lot more real estate out there for infringement to occur.”
What is the answer? Toerek suggests a due-diligence scrub of the video to catch unintentional infringement. Even if the infringement is “innocent and unintentional,” Toerek says it may still be actionable.
When I ask Toerek whether this legal thicket is discouraging brands from using AI, she answers flatly that it has not. So far, at least, companies consider the legal risks manageable in comparison to the significant cost savings of AI-assisted production.
At IdeaRocket, we’re excited by the rapid development of this technology. We think human creativity is still key, but these tools can amplify both our communication and our imagination.
