Artificial intelligence sits awkwardly inside a legal framework built for human creators, and the intellectual property issues around AI break into a handful of practical questions: who owns what AI helps make, whether training AI on copyrighted material is legal, whether AI can be named on a patent, and who pays when AI output infringes someone else’s rights. The answers are still developing, but enough is settled to guide anyone using AI commercially.
The short version: AI itself cannot be an author or inventor under U.S. law. Human contribution determines what’s protectable. And if you publish, sell, or distribute AI output that infringes, you are the one on the hook.
Can You Copyright Something AI Helped Create
Copyright protects “original works of authorship,” and every federal body that has weighed in agrees the author must be a human being.1Office of the Law Revision Counsel. 17 U.S. Code 102 – Subject Matter of Copyright In General The Copyright Office has stated that “the term ‘author’ excludes non-humans” and that its “registration policies have long required that works be the product of human authorship.”2Federal Register. Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence A work generated entirely by AI, with no meaningful human creative input, cannot be copyrighted, and the Supreme Court declined in March 2026 to disturb that rule after Stephen Thaler tried to register an image his “Creativity Machine” produced on its own.3U.S. Court of Appeals for the D.C. Circuit. Thaler v. Perlmutter
The harder question is what happens when a person uses AI as one tool in a larger creative process. The Copyright Office’s 2023 decision on the graphic novel Zarya of the Dawn drew the clearest line to date. Kris Kashtanova wrote the text, chose which Midjourney images to include, and arranged everything into a cohesive book. The Office protected her text and her creative selection and arrangement, but denied protection for the individual AI-generated images.4U.S. Copyright Office. Zarya of the Dawn Registration Decision Letter
The reasoning turned on control. A prompt describes what the author wants, but it doesn’t dictate the specific color of a character’s hands, the details of a photograph, or exactly how visual elements come together. Because of that gap, Kashtanova wasn’t the “master mind” behind the images. Small edits she made, like adjusting a character’s lip shape, were “too minor and imperceptible to supply the necessary creativity for copyright protection.”4U.S. Copyright Office. Zarya of the Dawn Registration Decision Letter
In January 2025, the Copyright Office issued a broader report confirming that existing law can handle these questions without new legislation. Key takeaways: prompts alone don’t create authorship, but a human who exercises “ultimate creative control” over the output can hold copyright in the original expression they contributed, and copyright can attach to the creative selection, coordination, or arrangement of AI-generated material even when the individual pieces are not protectable. The Office also concluded that “the case has not been made” for new legal protection specifically for AI-generated content, so material produced without meaningful human creative involvement enters the public domain.5U.S. Copyright Office. Copyright and Artificial Intelligence Part 2 Copyrightability Report
Can AI Be Listed as an Inventor on a Patent
No. The Patent Act defines an “inventor” as “the individual or, if a joint invention, the individuals collectively who invented or discovered the subject matter of the invention.”6Office of the Law Revision Counsel. 35 U.S. Code 100 – Definitions In Thaler v. Vidal, the Federal Circuit read “individual” to mean a natural person and rejected patent applications listing the AI system DABUS as inventor: “Congress has determined that only a natural person can be an inventor, so AI cannot be.”7U.S. Court of Appeals for the Federal Circuit. Thaler v. Vidal
That does not make AI-assisted inventions unpatentable. In February 2024, the USPTO issued guidance confirming that “AI-assisted inventions are not categorically unpatentable.”8United States Patent and Trademark Office. USPTO Issues Inventorship Guidance and Examples for AI-Assisted Inventions The question is whether a named human made a “significant contribution.”9Federal Register. Inventorship Guidance for AI-Assisted Inventions
Practically, that means:
- Identifying a problem and handing it to an AI is not enough to make you an inventor of whatever the AI produces.
- General oversight of an AI system, without more, does not qualify either.
- Designing a specific prompt that led to the invention can qualify, as can taking raw AI output and making meaningful technical contributions that refine it into something patentable.
Is It Legal to Train AI on Copyrighted Material
This is the most commercially consequential AI-IP fight active in the courts. Large language models and image generators are trained on massive datasets pulled from the internet, and those datasets inevitably include copyrighted articles, books, photographs, and code. Content creators call it unauthorized copying at scale. AI developers call it fair use.
Fair use permits limited use of copyrighted material without permission. Courts weigh four factors: the purpose and character of the use (including whether it’s “transformative”), the nature of the copyrighted work, how much was used, and the effect on the market for the original.10U.S. Copyright Office. Fair Use Index – About Fair Use AI companies lean on the first factor, arguing that training a model on text is different from reading or republishing it. Rights holders lean on the fourth, arguing that AI outputs now compete directly with the works that trained them.
Federal courts split during 2025. In Thomson Reuters v. Ross Intelligence, a Delaware federal court rejected the fair use defense, finding that Ross had used Thomson Reuters’ copyrighted legal headnotes to build a competing legal research tool. The court emphasized that Ross “was trying to build a substitute for Westlaw using Westlaw’s own curated content,” and weighed the effect on both the current market and a potential AI-training licensing market against fair use.11United States District Court for the District of Delaware. Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc. That ruling is on appeal.
Other courts went the other way. In separate cases involving Anthropic and Meta, federal judges found that training language models on copyrighted books was transformative fair use, largely because the models don’t reproduce or serve as substitutes for the original books. The pattern emerging from those decisions: using copyrighted material to build a direct competitor carries far more legal risk than using it to build something functionally different from the source material.
The highest-profile case, The New York Times v. Microsoft and OpenAI, is still being litigated in the Southern District of New York. The Times alleges that millions of its articles were used without permission to train models that now compete with its journalism.12United States District Court Southern District of New York. 23-cv-11195 The New York Times Company v. Microsoft Corporation et al. Its outcome will matter enormously to the direction of the law.
Trademark Risk From AI-Generated Branding
Under the Lanham Act, anyone who uses a mark in commerce in a way “likely to cause confusion” about the source, sponsorship, or affiliation of goods or services faces liability for infringement.13Office of the Law Revision Counsel. 15 U.S. Code 1125 – False Designations of Origin and False Descriptions Forbidden The standard doesn’t care whether a person or a machine produced the infringing material.
That creates a specific hazard for businesses using AI to build logos, marketing imagery, or product designs. AI image tools can produce output that closely resembles an existing trademark, and the user may not recognize the similarity. Intent is not a defense. If you put confusingly similar branding into commerce, you are exposed, regardless of how the image came to exist.
Right of Publicity and AI-Generated Likenesses
AI’s ability to generate realistic images, audio, and video of real people creates a category of risk that sits outside copyright and trademark. Generating a fake endorsement, placing someone’s likeness in a commercial context they didn’t authorize, or cloning a voice can violate right-of-publicity laws, which protect a person’s control over the commercial use of their name, image, and likeness.
Right of publicity is primarily state law, and the landscape is moving quickly. Several states have enacted or updated statutes in the past two years addressing AI-generated deepfakes and digital replicas, with coverage running from non-consensual intimate imagery to deceptive election content to unauthorized digital replicas of deceased performers. Penalties vary from state to state, with some jurisdictions imposing criminal liability and others offering only civil remedies.
At the federal level, the NO FAKES Act was introduced in 2024 to create a national framework against unauthorized AI replicas of a person’s voice and likeness.14U.S. Congress. S.4875 – NO FAKES Act of 2024 It was referred to the Senate Judiciary Committee and did not pass during that session. Until federal legislation lands, protection depends on which state’s law applies.
Who Pays When AI Output Infringes
This is where anyone using AI commercially should pay the closest attention. If you publish, sell, or distribute AI-generated content that infringes someone else’s copyright or trademark, you are the party facing legal exposure. Courts have shown no interest in treating “the AI did it” as a defense.
The financial exposure for copyright infringement alone can be severe. A copyright owner can elect statutory damages instead of proving actual losses, and the range runs from $750 to $30,000 per work infringed. Willful infringement pushes the ceiling to $150,000 per work.15Office of the Law Revision Counsel. 17 U.S. Code 504 – Remedies for Infringement Damages and Profits A business that uses AI to generate dozens of marketing images, product descriptions, or design elements could face claims on each individual work, plus attorney fees and injunctions forcing the material off the market.
Most AI platform terms of service push this risk onto the user. Read the indemnification clause of whatever tool you’re using; it almost certainly makes you responsible for how the output gets used. A few larger enterprise AI providers now offer limited indemnification for certain commercial uses, but the scope varies and typically doesn’t cover output you have modified.
Trade Secrets in AI Development
For many AI companies, the most valuable intellectual property never appears in a patent filing or copyright registration. Proprietary algorithms, model architectures, training methodologies, and curated datasets are often protected as trade secrets. Unlike patents, trade secrets carry no time limit and require no public disclosure, but protection lasts only as long as the information stays confidential and the company takes reasonable steps to guard it.
The federal Defend Trade Secrets Act gives companies a private right of action when trade secrets are stolen, with remedies including injunctions, actual damages, disgorgement of the thief’s profits, and exemplary damages up to double the compensatory award in cases of willful and malicious misappropriation.16Office of the Law Revision Counsel. 18 U.S. Code 1836 – Civil Proceedings
There is a built-in tension for AI companies. Open-source releases of model weights and training details conflict directly with trade secret doctrine: once information is voluntarily disclosed, it loses trade secret status permanently. Some companies split the difference by open-sourcing weights while keeping training data and fine-tuning processes confidential, but that boundary is difficult to hold.
Practical Steps to Reduce Your Exposure
The uncertainty in this area cuts both ways. It creates risk, but it also rewards anyone who documents their process carefully.
If you want copyright protection for AI-assisted work, the Copyright Office requires disclosure of AI use when more than a trivial amount of the output was generated by an AI system, along with a description of what the human author contributed. Failing to disclose can jeopardize the registration.5U.S. Copyright Office. Copyright and Artificial Intelligence Part 2 Copyrightability Report Keep records of your prompts, your iterations, and the creative decisions you made in selecting, arranging, or modifying AI output. That documentation is your evidence of human authorship if the question ever arises.
For patent applications involving AI-assisted inventions, document every step of your involvement. USPTO guidance requires a significant contribution beyond presenting a problem to an AI. Records of how you designed prompts, evaluated output, and made the technical decisions that shaped the invention support your inventorship claim.
Before using AI-generated content commercially, run trademark searches on any logos, brand elements, or names the AI produces. Check AI-generated images and text for similarity to known copyrighted works, particularly when the AI tool doesn’t disclose what it was trained on. And read the terms of service of every AI platform you use, with close attention to who owns the output and who carries liability if something goes wrong.