Artificial intelligence and copyright law are colliding in ways that affect real clients right now. Content creators want to know if their AI-assisted work is protected. Businesses want to know if using AI training data exposes them to liability. Lawyers want to know what to tell all of them. The law is not settled, but that does not mean you cannot give useful advice. This guide walks through the core questions: whether AI outputs can be copyrighted, whether training on protected works is lawful, who owns AI-assisted work product, and how to manage risk while doctrine continues to develop.
Why AI Has Made Copyright Questions More Urgent
Copyright law already addresses copying, derivative works, authorship, and fair use. Generative AI does not invent new legal categories, but it applies existing ones at a scale and speed that courts and agencies were not built to handle. A single AI system can ingest millions of books, images, or songs, generate new expressive content in seconds, and distribute it globally. That changes the practical stakes for law firms, companies, content creators, software developers, media organizations, courts, and government agencies.
The Shift From Human Tools to Autonomous Generation
A word processor helps a human write. A generative AI system can produce a complete article, image, or song from a short prompt with no further human input. That distinction matters because copyright doctrine has always assumed a human author making expressive choices. When a machine makes those choices autonomously, the traditional framework starts to strain.
Why Legal Professionals Need a Working Framework
Clients cannot wait for every question to be resolved. They need guidance now on risk allocation, contract terms, internal policies, documentation practices, and litigation exposure. The legal landscape is developing through agency guidance, court decisions, and private ordering. A working framework, even an imperfect one, is more useful than waiting for certainty that may not arrive for years.
Copyright Basics That Still Matter in the AI Era
Copyright protects original works of authorship fixed in a tangible medium. The owner holds exclusive rights to reproduce, distribute, display, perform, and prepare derivative works. Infringement occurs when someone exercises those rights without authorization or a valid defense. Fair use is the most important defense in AI disputes, but it is a fact-specific inquiry, not a safe harbor.
Originality and Human Authorship
Originality requires independent creation and at least a minimal degree of creativity. Courts and the Copyright Office have consistently held that human creative contribution is essential. Purely machine-generated output faces serious protection problems. AI-assisted work may qualify if a human contributes sufficient expressive choices, such as selecting, arranging, editing, or shaping the material.
The Copyright Office's January 2025 report reaffirmed this position. Prompts alone are not enough. The Office concluded that prompts "essentially function as instructions that convey unprotectible ideas" and do not give users authorship over the output. But where a human modifies AI-generated material to a meaningful degree, or where their own copyrightable expression is perceptible in the output, protection may attach to those human contributions.
Exclusive Rights and Potential AI-Related Infringement
AI training may implicate the reproduction right. AI outputs may implicate the derivative works right. Distribution of AI-generated content may implicate distribution and display rights. Which rights are at issue depends on the facts: what was copied, how, for what purpose, and what came out the other end.
Fair Use as a Central Battleground
Fair use turns on four factors: the purpose and character of the use, the nature of the copyrighted work, the amount used, and the effect on the market for the original. AI developers tend to emphasize transformativeness and the absence of market substitution. Copyright owners tend to emphasize commercial use, full copying, and lost licensing revenue. Both sides can point to plausible arguments, which is why litigation has exploded.
Can AI-Generated Works Be Copyrighted?
The short answer is: it depends on how much a human contributed. The Copyright Office and courts have drawn a line between fully machine-generated output and human-authored work that uses AI as a tool.
Fully Machine-Generated Output
If a user enters a short prompt and accepts the result without meaningful editing, the output is unlikely to qualify for copyright protection. The D.C. Circuit's ruling in Thaler v. Perlmutter made this clear: the Copyright Act requires human authorship. The Supreme Court denied certiorari in March 2026, leaving that ruling intact.
AI-Assisted Human Work
A lawyer who uses AI to generate a first draft, then substantially revises the structure, arguments, and language, may retain copyright in the resulting document. The same logic applies to a designer who uses AI to generate options and then selects, edits, and refines them. The key is whether identifiable human expression, selection, or arrangement is present in the final work.
Documenting Human Contribution
If copyright protection matters, document the process. Keep drafts, prompts, revision histories, and records of editorial decisions. This evidence can support registration, enforcement, licensing, and ownership disputes. The pending Allen v. Perlmutter case, involving an artist who used hundreds of iterative prompts to refine an AI-generated image, may eventually clarify how much human involvement is enough.
Is Training AI on Copyrighted Works Infringement?
This is the most contested question in AI copyright law. Training a generative AI model typically involves copying large volumes of text, images, or other content and using it to adjust the model's parameters. Copyright owners argue that copying is copying. AI developers argue that training is transformative analysis, not expressive reproduction.
The Copyright Owner's Argument
Rights holders argue that training involves unauthorized reproduction of protected works, extracts expressive value without compensation, and enables outputs that compete with or mimic the originals. They also argue that training undermines potential licensing markets, since developers could have paid for access but chose not to.
The AI Developer's Argument
Developers argue that training is transformative because the model learns patterns rather than reproducing expression. They compare it to a human reading widely to develop skills. They also argue that training outputs are not substitutes for the originals and that market harm is speculative.
Why the Answer May Depend on Facts and Markets
Courts are not treating this as an all-or-nothing question. In Thomson Reuters v. ROSS Intelligence, the court rejected fair use where the AI was trained on Westlaw headnotes to build a direct competitor. The court called market effect "the single most important element of fair use." By contrast, in Kadrey v. Meta, a court found that training on books was "highly transformative" where plaintiffs could not show concrete market harm. And in Bartz v. Anthropic, the same court drew a sharp line: training on lawfully acquired books was fair use, but maintaining a library of pirated copies was not. Anthropic later settled for $1.5 billion, the largest copyright settlement in U.S. history.
The pattern emerging from 2025 decisions: courts are increasingly receptive to fair use arguments for training on lawfully acquired data, deeply skeptical of speculative market-harm claims, and uniformly intolerant of piracy.
Who Owns AI-Assisted Work Product?
Ownership depends on copyrightability, employment status, contracts, platform terms, and the degree of human authorship. None of those factors operate in isolation.
Employees, Contractors, and Work Made for Hire
If an employee uses AI tools within the scope of their job, the work-made-for-hire doctrine should apply to any copyrightable elements, meaning the human-authored portions belong to the employer. But purely AI-generated portions may not be copyrightable at all, which means a work-made-for-hire clause alone may not resolve ownership. Organizations should add explicit assignment language for AI-assisted content, update employment agreements, and require disclosure of AI use in deliverables.
Client Work and Professional Responsibility
Lawyers using AI tools in client work face confidentiality obligations under Model Rule 1.6. Before uploading client documents to any AI platform, you need to understand how that platform handles data: what it stores, how long it retains it, and whether it uses inputs for model training. Our AI Disclosure and Trust Center address this directly. BriefCatch does not store document text, does not use client content for AI training, and processes text in RAM only. As our AI Disclosure states, "your data will never be used to improve or train AI models." That is the standard lawyers should apply when evaluating any tool they use on client matters.
Platform Terms and Licensing Restrictions
Copyright doctrine is only part of the answer. AI tool terms of service may restrict commercial use, impose attribution requirements, disclaim ownership of outputs, or grant the platform a license to use your inputs. Review those terms before approving any AI tool for client work or commercial materials. What the law allows and what a contract permits are different questions.
AI Outputs, Substantial Similarity, and Infringement Risk
Even if you did not intend to copy anything, an AI output might still infringe. Infringement analysis turns on access, substantial similarity, and whether the similarity involves protectable expression. Users who generate and distribute AI content can face liability even if the model, not the user, did the copying.
Prompting for a Specific Style or Existing Work
Asking an AI tool to write in the style of a living author, imitate a known character, or reproduce the feel of a specific copyrighted work increases both legal and reputational risk. Style alone is not copyrightable, but specific expression, plot elements, characters, and distinctive creative choices are. The closer the prompt pushes toward a specific protected work, the higher the risk that the output crosses the line.
Using AI Outputs in Commercial Settings
Marketing materials, product designs, software code, training manuals, legal documents, and public-facing content all carry higher stakes than internal experimentation. Before publishing or distributing AI-generated content commercially, run a review process that checks for similarity to known works, confirms platform terms permit commercial use, and documents any human creative contributions.
A Simple Risk Review Example
Say a company uses an AI tool to generate a product illustration. Before using it commercially, the team should ask: Does this resemble any known copyrighted image? What do the platform's terms say about commercial use? Did a human make meaningful creative choices in selecting or modifying the output? Is there documentation of those choices? That four-question check does not eliminate risk, but it creates a defensible record.
Practical Risk Management for Lawyers and Organizations
You do not need settled law to build a sensible governance framework. The following steps reduce copyright, confidentiality, and ownership risk while the doctrine continues to develop.
Create an Internal AI Use Policy
A policy should cover permitted tools, prohibited uses, approval workflows, confidentiality rules, review requirements, and documentation practices. Tailor it by practice area and risk tolerance. By 2025, 44% of law firms still lacked formal AI governance policies, which creates shadow AI risk where lawyers use unapproved tools without oversight. A blanket ban rarely works. A structured policy does.
Require Human Review and Verification
AI output should not go out the door unchecked. Review for accuracy, originality, citation reliability, legal reasoning, tone, and compliance with client instructions. Courts have sanctioned attorneys for submitting AI-hallucinated citations. In Wadsworth v. Walmart, 348 F.R.D. 489 (D. Wyo. 2025), eight of nine cited cases were fabricated by AI. FRCP 11 obligations do not change because the tool did the drafting. As our ethics guide puts it: "AI tools should streamline legal tasks, not replace your judgment."
Track Prompts, Inputs, and Revisions When Rights Matter
For any work that may be registered, licensed, enforced, or monetized, preserve evidence of human creative decisions. Save prompt sequences, drafts, revision histories, and records showing how a human shaped the final output. This documentation can make the difference between a protectable work and an unprotectable one.
Negotiate AI-Specific Contract Terms
Standard contracts were not written with generative AI in mind. When drafting or reviewing agreements, address representations and warranties about AI use, indemnity for infringing outputs, ownership of AI-assisted deliverables, confidentiality and data security obligations, restrictions on using client content for training, and disclosure requirements. These provisions matter whether you are the client, the vendor, or the lawyer advising either.
What Courts and Legal Writers Should Watch Next
The New York Times v. Microsoft/OpenAI litigation is in discovery and could produce significant rulings on news content, market substitution, and statutory damages. Music industry cases are resolving through settlement, with licensing agreements emerging as a practical alternative to litigation. Legislative proposals including the Generative AI Copyright Disclosure Act and the NO FAKES Act remain active. The Copyright Office's three-part report series, the third installment of which addressed AI training and fair use in May 2025, will continue to shape agency guidance even as political uncertainty surrounds the Office itself.
Key Questions Likely to Shape the Next Phase
- When does AI training qualify as fair use, and does the answer change based on the model's purpose or outputs?
- How much human input is enough to trigger copyright protection for AI-assisted works?
- Who bears liability for infringing outputs: the developer, the platform, or the user?
- How will licensing markets develop, and will compulsory licensing become a legislative option despite the Copyright Office recommending against it?
- What disclosure duties will courts, bar associations, or regulators impose on lawyers and businesses using AI?
Why Clear Legal Analysis Matters
Lawyers advising on AI and copyright need to be precise, balanced, and persuasive. Overclaiming in either direction, whether dismissing all risk or treating every AI use as infringement, does clients a disservice. The quality of the legal writing matters as much as the quality of the analysis. Briefs, memos, client alerts, and internal policies on AI topics need to hold up under scrutiny.
Writing About AI Copyright Issues With Precision
AI copyright analysis involves genuine uncertainty, competing analogies, and fast-moving developments. The goal is not to pretend you have answers you do not have. It is to communicate clearly about what is settled, what is contested, and what the risks are.
Use Clear Distinctions
Sloppy terminology creates analytical confusion. Keep these pairs separate: input versus output, training versus generation, copyrightability versus infringement, ownership versus license rights, and legal risk versus business risk. A client who understands the difference between "this output may not be copyrightable" and "this output may infringe someone else's copyright" is better positioned to make decisions.
State Uncertainty Without Sounding Vague
Precision and caution are not opposites. Formulations like "courts may consider," "risk increases when," or "the stronger argument is" give clients useful guidance without overclaiming. Avoid hedging so heavily that the advice becomes useless. A memo that says "it depends" without explaining what it depends on is not helpful.
Where BriefCatch Can Help
Legal professionals drafting client alerts, briefs, memos, internal policies, or court documents on AI copyright issues face a particular challenge: the subject is complex, the law is unsettled, and the stakes for getting the writing wrong are high. BriefCatch works inside Microsoft Word to provide real-time editing suggestions, citation guidance, and clarity-focused recommendations. It helps you sharpen the analysis you have already done, not replace the judgment behind it. If you are producing legal writing on fast-moving topics like AI and copyright, that kind of editing support can make a real difference.
A Practical Path Through a Moving Legal Landscape
Artificial intelligence and copyright law are not on a collision course that ends with one side winning. They are on a long adjustment path where doctrine, technology, markets, and contracts will all play a role. The fundamentals of copyright have not changed: human authorship matters, fair use is fact-specific, and market harm is often decisive. What has changed is the scale, speed, and complexity of the questions those fundamentals must answer.
Legal professionals who combine doctrinal knowledge with documented human judgment, sound governance, and clear writing are better positioned to advise clients, draft enforceable agreements, and produce persuasive arguments as the law develops. That combination is not a guarantee against uncertainty, but it is the most defensible approach available right now.
If you want help producing clearer, more precise legal analysis on AI copyright issues and everything else, try BriefCatch free or book a demo to see how it works in your practice.
.svg)
.svg)
.svg)
.svg)