Copyright, patent, and trade secret law are straining under questions they were never designed to answer.
The intellectual property system was built on a foundational assumption so obvious that lawmakers never thought to write it down: creators are human.
That assumption is now under siege.
Artificial intelligence is generating music, drafting patents, authoring code, designing drugs, and producing visual art at a scale and speed no human can match. Our IP frameworks (copyright, patent, trade secret) are straining under the weight of questions they were never designed to answer.
The authorship problem
Copyright law, at its core, rewards creative expression. The Supreme Court has long held that copyright protection requires human authorship. The Copyright Office has reaffirmed this repeatedly in recent guidance: works produced entirely by a machine, without creative input from a human, are not eligible for protection.
But here is where it gets complicated. When a human uses an AI tool to generate an image, write a song, or draft a novel, where does the machine's contribution end and the human's begin? The spectrum runs from a human who typed a three-word prompt to one who spent weeks iterating, curating, and shaping AI output. The law gives us a binary (protected or not), while reality gives us a gradient.
The result is legal uncertainty that chills both innovation and investment. Creators don't know what they own. Businesses don't know what they can protect. And the courts are just beginning to grapple with cases that will define the answer for a generation.
The inventorship problem
Patent law has a parallel crisis. The U.S. Patent Act requires that inventors be natural persons. When a pharmaceutical company uses AI to identify a novel drug candidate, or an engineering firm uses AI to design a new mechanical component, who is the inventor? The human who built the AI? The human who ran the query? The AI itself?
Courts in the United States, the United Kingdom, and Australia have now weighed in, uniformly holding that AI cannot be a named inventor. The Federal Circuit's decision in Thaler v. Vidal was unambiguous: inventors must be natural persons.
This creates a perverse dynamic. Companies are quietly filing patents on AI-generated inventions, listing human employees as inventors; people who may have contributed little more than pressing "run." The integrity of the inventorship declaration is now in serious jeopardy. We are building a patent system increasingly populated by legal fictions.
The training data problem
Perhaps the deepest tension is one that precedes all of the above: the AI systems generating these works were trained on copyrighted material (books, articles, images, code), often scraped without permission, license, or compensation to the original creators.
A wave of litigation is now testing whether training an AI on copyrighted works constitutes infringement. If training is infringement, the current generation of large language models rests on an unstable legal foundation. If training is fair use, we have effectively decided that creators bear the cost of building the AI systems that may ultimately displace them.
Neither outcome is fully satisfying. What we need is a licensing framework, one that compensates creators while allowing the innovation ecosystem to function. We have built such frameworks before, for music sampling, for cable television, for digital audio recording. We can do it again.
The incentive mismatch
Underlying all of these disputes is a deeper structural problem. IP law is premised on an incentive theory: we grant temporary monopolies to encourage investment in creative and inventive activity. The logic assumes that without IP protection, creators would underinvest.
AI scrambles this logic in multiple directions at once. On one hand, AI dramatically lowers the cost of generating output. If a generative model can produce ten thousand images in an hour, the traditional scarcity argument for copyright weakens considerably. On the other hand, training and deploying frontier AI systems requires enormous investment, and the companies making those investments argue they need IP protection to recoup costs.
The question worth asking is: who are we trying to incentivize? If the answer is human creativity, then IP protections for AI-generated works may be exactly backwards. If the answer is investment in AI infrastructure, we should say so clearly, rather than laundering AI-generated content through human-authorship fictions.
The path forward
Several steps are worth serious consideration:
- Transparency requirements. AI-generated or AI-assisted works should be disclosed in patent applications and copyright registrations. Legal fictions serve no one.
- A compulsory licensing regime for training data. Creators whose work was used to train AI systems deserve compensation. A collective licensing structure, similar to those used in music, could provide it without halting AI development.
- A new category of protection, or none at all. Rather than forcing AI-generated works into frameworks designed for human creators, Congress should consider a new, more limited form of protection, or whether the public domain is the right default for machine output.
- Ongoing review. IP policy has historically moved slowly. AI is moving fast. We need mechanisms for regular reassessment, not one-time legislation that will be obsolete before the ink is dry.
The choices we make in the next few years will shape the relationship between human creativity and machine intelligence for decades. Getting them right matters, not just for lawyers and technologists, but for every artist, scientist, and inventor whose work will be touched by these systems.
The law has always had to adapt to new technologies. The question is whether we adapt it thoughtfully, or let the machines adapt it for us.