When users ask ChatGPT to write a short story, create a graphic, or compose a song, the LLM produces something that looks like an original work within seconds. But who owns the copyright to the work products AI models create? And what happens when the LLM uses someone else’s intellectual property to produce it?
As artificial intelligence becomes an increasingly common tool, these questions are challenging longstanding pillars of copyright and intellectual property law. According to Fordham Law professors Joseph Fishman and Ron Lazebnik, the answers aren’t always straightforward.
“As new technologies come up, it raises thornier questions,” said Lazebnik, who specializes in intellectual property and the intersection of law and technology. “We are still very early on in terms of the courts analyzing these issues.”
Can You Copyright Something Created by AI?
Under current U.S. Copyright Office guidance, works created entirely by AI aren’t eligible for copyright. That’s because current law requires human authorship, according to Fishman, whose work focuses on intellectual property and the music industry.
That means that if you ask an AI tool to write a poem or generate an illustration, you typically can’t claim exclusive ownership of the result. “In most scenarios, nobody owns the final product,” said Fishman, Fordham Law’s Bacon-Kilkenny Distinguished Research Scholar.
But what if you spend hours refining prompts to get exactly the result you want? The Copyright Office currently maintains that prompting alone, regardless of how extensive, isn’t enough to establish authorship, Fishman said. That position is currently being challenged in court.
Is Training AI on Copyrighted Material Legal?
AI models learn from enormous datasets of text, images, music, and other materials. Whether companies can legally use copyrighted works to train their models is one of the biggest unresolved questions surrounding AI and intellectual property.
The debate hinges on fair use, a legal doctrine that permits certain uses of copyrighted works without the owner’s permission. Courts consider factors like the purpose of the use, how much material was copied, and whether the new use harms the market for the original work, according to Lazebnik.
“The question framed by some is, assuming you legitimately accessed the material, are you now doing something that is considered fair with it, or not?” he said.
He pointed to litigation involving AI company Anthropic, in which a federal judge distinguished between training its model Claude on lawfully acquired books versus using pirated copies, finding that using legally acquired books constituted fair use.
But Lazebnik cautioned that the ruling isn’t necessarily settled law. “It’s not 100% clear to me that this is where the law will ultimately rest,” he said.
Can You Get Sued for Using AI-Generated Content?
Potentially, yes, according to Fishman. Consider a business that uses AI to create a commercial jingle. If the result is substantially similar to an existing copyrighted song, the business could face liability, even if nobody realized the AI had reproduced protected material, Fishman said.
“Their intent may be innocent, and that might mitigate the damages they would owe, but it may not save them from liability for infringing on copyright,” Fishman said.
This matters for companies using AI tools to produce employee work products. An employer’s ownership of AI-assisted work may depend on the extent of the employee’s contributions.
Will Artists and Writers Be Compensated for AI Training?
What about all the creators whose work is being used to train AI models? Some AI companies are already negotiating licensing agreements with copyright holders, paying for access to works used in training. But Fishman says the difficult question is how much of that money will ultimately reach creators.
“If a model is trained on millions of copyrighted works, how much would make it back to the individual artist? There’s no obvious way to do that,” he said.
Still, the current case law could bode well for the future of creative workers, Fishman noted. If only human-generated work can be protected by copyright, then record labels, publishing houses, and other industries have strong incentives to continue relying on human labor.
For now, the copyright issues raised by AI leave legal experts, creative workers, and everyday users in a place of uncertainty, waiting on the courts for more definitive answers.
