🎧 Listen: https://unscripted-small-business.castos.com/episodes/ai-cant-be-an-author-or-an-inventor-ip-for-small-business | 📺 Watch: https://www.youtube.com/watch?v=4ZlEUdtLPJ4
I came into this one with a selfish question. I put an episode out every week, it gets syndicated to Spotify, Amazon Music, Deezer and a stack of RSS directories, most of them carrying a raw transcript, and I had never actually checked what protects any of it. Richard Gearhart has spent twenty years answering that question for other people. He founded Gearhart Law in Summit, New Jersey, the firm covers patents, trademarks, copyrights and trade secrets with about 25 contributors, and this month it is closing on its 5,000th client. What I got back was a lot bigger than podcasting. If your business makes something, writes something, or keeps something quiet, and you have let AI into any part of that workflow, a few things you assume are protected are not. Here is what actually holds up.
Four tools, four jobs
Richard opened by sorting the whole field into four buckets, and it is worth memorizing, because owners mix them up constantly.
“Patents protect technologies, trademarks protect brands, copyrights protect original works of expression like movies, novels, etc. And trade secrets protect secrets that give you a competitive advantage.”
If you cannot say which of those four you are relying on for a given asset, there is a decent chance you are relying on none of them.
Copyright starts the second you make the thing
Podcasting, he said, “is most likely to be governed by copyright law,” sitting in “the same category as a book or a movie.” Nothing to file. Nothing to pay.
“You automatically get copyright protection just because you made it, which is great. It’s like the best deal in intellectual property law ever.”
You can also file a copyright registration for the audio with the U.S. Library of Congress. You do not have to. That optional step turns out to be the most consequential decision in this whole article, which is the next section.
Registration is what makes it worth suing over
Early in my career I worked at a real estate hosting company where realtors uploaded photos to their own sites. One day a demand letter arrived from Getty’s lawyers. Five hundred dollars an image, or court. I filed it under shakedown and moved on. Richard represents photographers in that space and businesses on the receiving end of those letters, and he explained why the math works.
Damages in a copyright case are genuinely hard to prove. Registration deletes that problem.
“There’s a provision in the copyright law that if the work is registered, then they can get statutory damages, which pretty much means if you can prove infringement the damages are automatic and you don’t have to prove the damages.”
“And the law provides $500 to $20,000 per occurrence of infringement.”
That is the whole business model, and he was blunt about it: “Some photographers make more money doing that than they do selling their work.” He also refused to hand me a villain. “So what is morally right, I think both sides have legitimate arguments.” A photographer spends years learning a craft, and as he put it, “it’s not just like snapping photographs and posting them on Instagram, right? It’s a trade.”
The owner takeaway is two sentences long. Register the work you would actually go to court over. Never use an image you did not license, because the bill is real and it is calculated per occurrence.
Artificial intelligence cannot be an author
“Artificial intelligence cannot be an author.”
He used Google Notebook as the example, a tool that “can create podcasts that sound like a real podcast but it’s really just an AI voice.” There is no human behind that output, so there is nothing to own.
“That material would not be copyrightable, because there is no author other than the AI.”
Mixed work splits down the middle.
“Now, if it’s a mixed situation where you have a human being and AI together, then the parts of the podcast generated by AI would not be copyrightable, but the parts generated by the human would be copyrightable.”
Then comes the part nobody has settled. If a model writes the questions and a human asks them out loud, the spoken performance is protectable. The transcript is a genuine unknown. “Nobody really knows, right?” Heavy human direction in the prompting can put copyrightable material back into the mix, because at that point the human used the model as a tool. Where any specific line falls gets decided in a courtroom, one expert against another, “and then a judge and a jury listen to it and they make some sort of decision based on who knows what.”
On the platform is not the same as off the platform
“Once it goes on a social media platform, anybody can use it for anything if it’s on that social media platform.”
Facebook and YouTube terms hand out that reuse right, and Richard counts YouTube as social “because the companies want you to share.” The boundary is the platform edge.
“But if they take that content and post it on their website and it’s no longer on the social media platform, then that could be a copyright violation.”
Crediting you does not cure it: “but once you go outside the platform, attribution is technically not enough.” I told him I have the opposite hope for this show. I want people lifting pieces of it. “I hope somebody listens to it and takes a piece of it and puts it out there and then gives me an attribution.” That is entirely allowed, because the rights are mine and I get to not enforce them.
Which is his real point. His LeBron James example and the Oprah Winfrey one both land in the same place: what happens next depends on the rights holder’s appetite, not just the statute. The Oprah Winfrey Fan Club podcast said nothing but nice things about her and was still asked to stop using the name.
A trade secret stops being secret the moment AI can guess it
This is the section I would print out and tape to the wall if I manufactured anything. I work with an industrial manufacturer that makes oxygen monitors, so I asked Richard how defensible a secret process really is now that a competitor can throw guesses at a model until something sticks. Ask it what is in Coke, try the answer, and see.
“In general, if something is reverse engineerable to the public, then it’s not really a trade secret anymore.”
Traditional trade secrets are customer lists, manufacturing processes, recipes, “that old secret family recipe.” All of them depend on the thing staying unguessable, and that is exactly the assumption AI is eroding.
“So to the extent that you could use artificial intelligence to reverse engineer a trade secret, just based on publicly available information, then that weakens the trade secret protection substantially, even if the owners did everything they were supposed to do.”
Read that last clause twice. You can do every single thing right and still watch the protection thin out. His software example is the one that should worry founders. Companies leaned on trade secret protection because a front end never revealed the back end, and the customer could not infer the programming from the output.
“And with AI though, if AI could come up with a way to replicate what the customer is getting without actually doing the programming, then that would weaken the trade secret protection of the software company.”
The only comfort he offered is a temporary one. “AI still has a lot of problems, a lot of hallucinations, where it thinks it solved the problem but it hasn’t yet.”
AI cannot be an inventor either
I asked the version of this that matters to anyone building software with Claude Code or ChatGPT in the loop. Does using AI to write the thing cost you the patent?
“The patent laws require a human inventor.”
A case settled several years ago where the filing named only an AI system as the inventor, and the court said no. Gearhart Law wrote up that holding: AI cannot be an inventor.
Using the tool is still fine. “The patent office guidelines specifically allow you to use AI as a tool.” In his framing it is a lab technique, nothing more: “It’s no different from distilling a compound than it is using AI to draft code.”
The trap is not the tool. The trap is scope.
“The complexity comes in when AI starts adding ideas to the invention that the inventor never contemplated.”
The model suggests something, you like it, you put it in the filing, and “then in theory that part of the patent would not be enforceable.” So the drafting job changed.
“You just have to be careful that when you file the patent, you write it in a way so that the scope is really limited to what humans actually contributed to or control.”
And do not walk into his office with a machine-drafted application. “We have inventors that will have AI write the patents for them and they’ll bring them in, and they’re almost always wrong.” He expects that to change. “I do think within a couple of years, soonest, that AI will be able to draft good quality patents.” Not yet.
Putting your invention into a public model is publishing it
“When you put your ideas into an LLM, if it’s a public LLM, then technically you’re disclosing your invention.”
That one sentence should change how your product team uses a chat window, because disclosure has consequences and, as he said, people do not appreciate that.
It gets worse once lawyers are involved. He described a white collar defendant who fed his own case into a public model. The other side asked for all of it, the court agreed none of it was privileged, and when they read the model’s privacy policy it “had holes in it large enough to drive a truck through.” Some of what he had to turn over was incriminating.
A separate case went the same way. An expert wrote an opinion using models almost exclusively, it was turned over in discovery, and that “just killed the case for him right there, right then and there.”
His advice for owners is not to go become an amateur IP lawyer. “You shouldn’t be walking around thinking about intellectual property law unless you love it. It’s better to find somebody that you can trust and get your questions answered.”
What a 25 person law firm actually automates
The last question turned the mirror around, because this is Unscripted Small Business. I mentioned a lead response study by Ignitvio that looked at 1,824 businesses and found 68 percent never responded to somebody reaching out within twenty-four hours, and asked what his own stack looks like.
An outside company builds their systems, and confidentiality drives every choice in it.
“So there isn’t any AI system except for Microsoft Copilot that has exposure to our clients’ information.”
Copilot runs in the firm’s own tenant, and as he understands it that use is approved by the Bar Association. The workflow they actually automated is the boring one, which is usually the right answer. When correspondence comes in from the patent office, a system analyzes it, summarizes the response, calls out the deadlines and drafts the reporting letter to the client. An attorney reviews it before it goes anywhere. Legal research runs on Lexis.
He will not let it near his own email.
“I personally don’t use AI to write emails, because I like people hearing me.”
“I don’t want to lose my voice.”
And the line he draws is the one most owners are looking for: “That’s where I think AI can take some of the burden off of routine tasks, but for the higher level stuff you still need a person.”
Where to find Richard Gearhart
Richard Gearhart, Esq. is the founding partner of Gearhart Law, LLC in Summit, New Jersey, an intellectual property firm just past its twentieth anniversary and closing on its 5,000th client. Before founding the firm he ran the US Patent and Trademark group at Novartis. He also hosts the Passage to Profit Show. Find him on LinkedIn.
His closing offer on the episode was open ended: “If anybody is interested in intellectual property, knowing more, or if you have an issue, we’re always happy to discuss that.”
Six steps to audit what your team pastes into public AI tools, plus a printable one-page checklist. No email required.
Listen to the full episode
There is more on the recording than fits here, including the Anthropic pirated books lawsuit, whether running a photo through Midjourney gives you any cover, and his description of how a substantial similarity verdict really gets decided, which “maybe depends on the tie that the lawyer is wearing that day.”
🎧 Listen: https://unscripted-small-business.castos.com/episodes/ai-cant-be-an-author-or-an-inventor-ip-for-small-business | 📺 Watch: https://www.youtube.com/watch?v=4ZlEUdtLPJ4
Subscribe to Unscripted Small Business wherever you get your podcasts. If this one saved you a phone call, leave a rating. It is how other owners find the show.
Keep going → Browse the interviews