Artificial Intelligence and Copyright: A Practical Overview for Legal Professionals

Generative AI writes marketing copy, draws logos, and composes music in seconds. Copyright law was not built for that speed. Three rules now govern the collision. Copyright protects human expression, never machine output standing alone. Training AI on copyrighted works occupies genuinely unsettled legal ground. And when AI output infringes, the people and companies who used the tool can pay — not just the ones who built it.

This guide explains each rule with the statutes, cases, and Copyright Office positions behind it. It then gives you a checklist you can apply to your own work today, whether you create content in Missouri or advise clients who do.

Copyright protects human expression — never ideas, facts, or machines

Start with the statute. Under 17 U.S.C. § 102, copyright covers original works of authorship once the author fixes them in a tangible form. Section 102(b) withholds protection from any idea, procedure, process, system, or method of operation. The Supreme Court drove the idea-expression line home in Baker v. Selden, 101 U.S. 99 (1879), which refused copyright for an accounting system itself.

Originality sets a low but real bar. In Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991), the Court held that a work needs independent creation plus a modicum of creativity. Facts themselves stay free for everyone. A phone book's alphabetical listing failed the test; a creative selection or arrangement of facts can pass it.

Small building blocks stay free too. The Copyright Office's regulation at 37 C.F.R. § 202.1 keeps words, short phrases, names, and titles outside copyright entirely. These limits matter for AI. Much of what a model produces — stock phrases, common plots, generic chord progressions — was never protected in the first place.

Ownership also brings a defined bundle of rights. Section 106 gives the copyright owner the exclusive power to reproduce the work, prepare derivative works, distribute copies, and perform or display the work publicly. Every infringement question in this guide reduces to one inquiry: did someone exercise one of those rights without permission or a valid defense?

The human-authorship rule decides who can own AI output

American copyright runs on human effort. The Supreme Court treated a photographer as an author in Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53 (1884), because his creative choices — pose, lighting, framing — produced the picture. The camera was his tool. That tool logic has one hard edge: the author must be human.

The Ninth Circuit proved the edge in Naruto v. Slater, 888 F.3d 418 (9th Cir. 2018), the monkey-selfie case. A macaque pressed the shutter; the court held the macaque could not hold a copyright, because the Copyright Act confers rights on persons. The Copyright Office reads the Act the same way and refuses registration when a machine, not a human, did the expressive work.

The D.C. Circuit settled the question for AI in Thaler v. Perlmutter, 130 F.4th 1039 (D.C. Cir. 2025). Stephen Thaler sought registration for an image he attributed entirely to his AI system. The court affirmed the Copyright Office's refusal: the Copyright Act requires a human author, full stop. A work with no human creative contribution belongs to no one. It enters the public domain the moment it exists.

The Copyright Office's 2023–2025 framework: disclose the AI, claim the human parts

Most real projects mix human and machine effort. The Copyright Office built its framework for that reality. Its March 2023 registration guidance requires applicants to disclose AI-generated content in any work they register. The registration then covers only the human-authored contribution. Failing to disclose can sink the registration.

The Office's January 2025 report on copyrightability, part of its Copyright and Artificial Intelligence initiative, refined the test. Prompts alone usually do not make you an author, because you do not control the expressive details the model produces. But human selection, arrangement, and modification of AI output can earn protection for those human contributions. The Office judges each application on its facts.

Two practical consequences follow. First, your protection covers your human layer — your edits, your arrangement, your original text woven through the output — not the raw AI material. Second, your records matter. If you cannot show what you contributed, you cannot register it or enforce it.

Who owns AI-assisted work in practice

Ownership follows authorship. If your employee uses AI to draft a report within the scope of the job, the work-made-for-hire doctrine under 17 U.S.C. § 201(b) vests the human-authored elements in the employer. Freelance and vendor relationships need more care, because commissioned works become works made for hire only in defined categories and only with a signed writing.

Read your AI tool's terms of service. Many platforms assign you whatever rights they can give in the output — but they cannot grant copyright in material that has no human author. A subscription buys a license, not authorship. If a competitor copies your purely AI-generated image, you may find no copyright to enforce at all.

The fix is human craft. Draft the brief, direct the structure, rewrite the passages, and document what you did. The more of you in the final work, the more of it you own.

Training on copyrighted works is unsettled — plan for both outcomes

AI developers trained their models on enormous collections of books, articles, images, and songs. Rights holders answered with dozens of federal lawsuits against the major developers, and those cases remain pending. No appellate court has issued a definitive ruling on whether training is fair use.

Fair use under 17 U.S.C. § 107 weighs four factors: the purpose and character of the use, the nature of the copyrighted work, the amount taken, and the effect on the market for the original. Developers argue that training is transformative — the model learns patterns rather than republishing works. Rights holders answer that training copies everything and that AI output now competes with the very works it absorbed.

The Copyright Office's May 2025 report on generative AI training analyzed both sides and deliberately left the question to the courts. It flagged realistic licensing markets as a key variable and warned that market dilution cuts against fair use. Plan accordingly. If you license content, watch the case law. If you build on AI, know the legal ground under your training data could shift.

Early district-court activity points in more than one direction. Some judges have signaled sympathy for transformative-use arguments; others have let claims proceed on market-harm theories, and a few defendants have chosen to settle rather than litigate to judgment. Until an appellate court weighs in, treat every confident prediction you read with suspicion — including confident predictions from AI vendors.

When AI output infringes, users and developers both face exposure

Infringement under 17 U.S.C. § 501 occurs when someone exercises an exclusive right — reproduction, adaptation, distribution, public display — without permission. An AI image substantially similar to a protected photograph can infringe just as a hand-traced copy would. The machine's involvement is no defense.

The remedies carry weight. Section 502 authorizes injunctions. Section 504 allows actual damages plus the infringer's profits, or statutory damages of $750 to $30,000 per work — rising to $150,000 when the infringement is willful. Section 505 lets the court award costs and attorney's fees. Remember the gatekeeper rule of § 411(a): you generally must register your work before you can sue to enforce it.

Who pays when output infringes? The user who published it is the obvious defendant. Developers face secondary-liability theories when they knowingly facilitate infringement. No controlling precedent allocates responsibility between them yet. Courts will borrow from the secondary-liability tradition of Sony Corp. v. Universal City Studios, Inc., 464 U.S. 417 (1984), which shields technologies with substantial non-infringing uses, and from later inducement doctrine. Until clearer law arrives, treat output review as your job, not the vendor's.

What this means for Missouri readers

Copyright is federal law, and it preempts any Missouri state-law equivalent under 17 U.S.C. § 301. A copyright suit from St. Louis goes to the U.S. District Court for the Eastern District of Missouri; one from Kansas City goes to the Western District. Appeals land in the Eighth Circuit. No Missouri or Eighth Circuit decision yet addresses AI authorship or training, so the national authorities above control the advice you will hear here.

The exposure is local even if the law is federal. Missouri marketing agencies, musicians, architects, and small publishers now use generative tools daily. Their questions are the same ones this guide answers: Can I register this? Could this image infringe? What do I tell clients? Federal law supplies the answers; Missouri businesses must still act on them.

What to do now: a five-step checklist

• Log your process. Keep prompts, drafts, and edits for anything you may register or enforce. Your records prove your human contribution.
• Disclose AI content on every registration application, and claim only the human-authored material. The 2023 guidance requires it.
• Put copyright terms in writing with employees, freelancers, and AI vendors. Do not assume a subscription or a handshake settles ownership.
• Review AI output against known sources before you publish it. Substantial similarity liability does not care that a machine drafted the work.
• Register your important human-authored works early. Registration unlocks the courthouse under § 411(a) and statutory damages under § 504.

Common questions about AI and copyright

Can I register a copyright in something an AI wrote for me?

Not for the machine's raw output. You can register your own contributions — original text you added, creative arrangement, meaningful edits — if you disclose the AI content on the application.

Is it legal for my business to use generative AI tools?

Yes. No statute bans the tools. Your risk lies in how you use them: publishing infringing output, overclaiming ownership, or breaching a license. The checklist above manages each risk.

What if an AI image closely resembles my copyrighted photograph?

You may have an infringement claim if the output copies protectable expression from your work and the developer or user had access to it. Registration status and proof of your original elements will shape the claim, so talk to counsel before you act.

Does paying for an AI subscription make me the author of the output?

No. Payment buys contractual rights between you and the platform. Authorship still requires your human creative contribution under Thaler and the Copyright Office's framework.

Talk with a Missouri intellectual property attorney

AI copyright questions rarely arrive with clean facts. If you need to register a hybrid work, respond to a takedown demand, or set policy for a creative team, Ott Law Firm's intellectual property practice can help you work through it. Call (314) 710-2740 or contact the firm online to schedule a consultation. You can also browse more analysis on our blog.

This article is for informational purposes only and does not constitute legal advice. Reading it does not create an attorney-client relationship, and every matter turns on its own facts.