AI and Copyright Law: Who Owns AI-Generated Content?

In February 2023, the US Copyright Office made a decision that reverberated through the creative industry: it refused to register copyright for a comic book whose images were generated by Midjourney, the AI image tool. The author, Kristina Kashtanova, retained copyright over the text she wrote — but the AI-generated artwork was deemed unprotectable under US law.

That single ruling crystallized a legal crisis that has since spawned dozens of lawsuits, regulatory proceedings, and international policy debates. The central question: in an age of generative AI, who owns what?

The Human Authorship Requirement

US copyright law has long required a human author. The Copyright Act of 1976 protects ‘original works of authorship’ — and both courts and the Copyright Office have interpreted ‘authorship’ to require human creative expression.

The foundational precedent comes from an 1884 Supreme Court case, Burrow-Giles Lithographic Co. v. Sarony, which held that photography could be copyrighted because it involved human creative choices. Courts have since applied that principle to reject copyright claims for works created without sufficient human creativity.

In the 2023 case Thaler v. Perlmutter (DC District Court, August 2023), Judge Beryl Howell ruled that AI-generated art cannot be copyrighted because copyright ‘has never stretched so far’ as to protect work created by non-humans. Stephen Thaler, who had attempted to register artwork created by his ‘DABUS’ AI system, lost his appeal to the DC Circuit in 2024.

The Copyright Office’s Evolving Guidance

The US Copyright Office has issued several key policy statements:

• February 2023: Copyright in AI-generated images requires ‘sufficient human authorship’ — pure AI output is unprotectable
• March 2023: Guidance clarifying that works containing AI-generated material must disclose that fact in registration applications
• August 2023: Copyright Office launched a major study on AI and copyright, soliciting public comments (over 10,000 received)
• May 2024: Part 1 of the Copyright and Artificial Intelligence Report published, addressing digital replicas (deepfakes)
• July 2024: Part 2 addressing AI training and fair use

The Office has taken the position that the key question is whether a human made ‘sufficient creative choices’ in the work. If an artist uses AI as a tool — selecting outputs, arranging elements, modifying results — they may retain copyright over the final composition. Pure prompt-to-output generation with minimal human selection is unlikely to qualify.

The Training Data Lawsuits

A parallel legal battle is unfolding over whether AI companies infringed copyright by training their models on copyrighted works scraped from the internet — without permission, compensation, or attribution.

Key Lawsuits

• Getty Images v. Stability AI (filed January 2023, UK and US): Getty claims Stability AI scraped over 12 million copyrighted images. The US case (Delaware District Court) survived a motion to dismiss in February 2024.
• Authors Guild v. OpenAI (filed September 2023): Authors including John Grisham, Jodi Picoult, and George R.R. Martin allege OpenAI trained on their books without authorization.
• New York Times v. OpenAI and Microsoft (filed December 2023): The NYT claims its articles were used to train GPT models; OpenAI’s output sometimes reproduces NYT articles nearly verbatim.
• Concord Music Group v. Anthropic (filed October 2023): Music publishers allege Claude reproduces song lyrics, which are copyrighted separately from recordings.
• Visual Artists Guild v. Stability AI, DeviantArt, Midjourney: Class action by artists claiming image generators are ‘collages’ built from their work.

As of early 2026, most cases remain in discovery or early procedural stages. The outcomes will fundamentally shape whether generative AI’s training process constitutes copyright infringement — a question worth hundreds of billions of dollars.

The Fair Use Defense

AI companies have consistently argued that training on copyrighted data constitutes fair use under 17 U.S.C. § 107. Fair use analysis involves four factors:

• Purpose and character of use: Is it transformative? AI training involves learning statistical patterns, not reproducing the work — potentially transformative
• Nature of copyrighted work: Published works receive less protection than unpublished ones
• Amount used: Training typically uses entire works, which typically weighs against fair use
• Effect on the market: Does AI output substitute for the original? This is the most contested factor

The closest precedent is Authors Guild v. Google (2d Cir. 2015), which found Google’s scanning of books for its search index was fair use because it was transformative (enabling search, not reproducing books). AI companies argue training is similarly transformative. Copyright holders argue AI outputs directly compete with and replace original works.

International Perspectives

European Union

The EU AI Act includes a copyright provision: GPAI model providers must comply with EU copyright law and publish ‘sufficiently detailed summaries’ of training data. The EU Copyright Directive (2019/790) introduced a text and data mining exception, but with an opt-out mechanism for rights holders — meaning copyright owners can legally prohibit their content from being used in AI training.

Japan

Japan has taken the most permissive approach globally. In May 2023, Japan’s Agency for Cultural Affairs confirmed that using copyrighted material for AI training does not infringe copyright under Japanese law — regardless of whether the material was obtained legally. This has made Japan attractive for AI research and training data collection.

United Kingdom

The UK attempted to introduce a broad text and data mining exception similar to Japan’s but withdrew it in 2023 following creative industry opposition. The UK Intellectual Property Office continues to consult on the issue.

What You Can and Cannot Protect

Given current law (primarily US-focused), here’s a practical breakdown:

• Can be protected: Your written prompts (if sufficiently creative and lengthy); human-written text with AI assistance; final works where you made substantial creative selections from AI outputs; arrangements and compilations of AI content
• Cannot be protected: Pure AI output generated with a simple prompt; AI-generated images you selected but did not modify; content where AI made all creative decisions

The Compensation Question

Beyond copyright, a broader question is emerging: should AI companies pay content creators for training data, even if training is ultimately deemed legal? In 2023, Adobe began offering compensation to stock photo contributors whose images are used in AI training. Shutterstock partnered with OpenAI under a licensing agreement. These voluntary deals hint at a potential market-based solution that could coexist with (or replace) litigation.

The Screen Actors Guild-American Federation of Television and Radio Artists (SAG-AFTRA) struck a deal in 2024 including AI protections for members’ likenesses and voices — the first major union contract to directly address generative AI training and synthetic media.

Frequently Asked Questions

Can I copyright content created with ChatGPT or Midjourney?

Not if the AI did all the creative work. US law requires human authorship. However, if you made substantial creative selections, arrangements, or modifications, the human-authored portions may be protectable. You must disclose AI use when registering with the US Copyright Office.

Who owns the copyright to AI output — the user or the AI company?

Under current law, no one automatically owns copyright to pure AI output in the US — it falls into the public domain. AI company terms of service may assign rights to you contractually, but that’s a contractual right, not a copyright. OpenAI, for example, assigns output ownership to users in its ToS, but courts haven’t definitively ruled this creates enforceable copyright.

Is it legal for AI companies to train on copyrighted content?

This is the central unsettled legal question. AI companies argue it’s fair use; copyright holders disagree. Multiple major lawsuits are pending. Outcomes will likely vary by jurisdiction and the specific facts of each case.

What is the EU copyright opt-out for AI training?

Under the EU Copyright Directive, rights holders can opt out of having their content used for AI training by publishing machine-readable notices. The EU AI Act reinforces this by requiring GPAI providers to comply with EU copyright law and respect opt-out notices.

Do I need to disclose if I used AI to create content I’m publishing?

Legally, it depends on jurisdiction and context. The US Copyright Office requires disclosure for registration purposes. Some platforms (news organizations, academic journals) have their own disclosure policies. The EU AI Act requires transparency for AI-generated content in some contexts.

Sources

• Grokipedia
• Stanford HAI AI Index Report

Related reading: AI Ethics for Beginners | AI Content Creation | AI Image Generation | Is AI Safe? | AI for Writers

Sources: US Copyright Office reports (2023–2024), court filings in Thaler v. Perlmutter, NYT v. OpenAI, Getty Images v. Stability AI; EU Copyright Directive 2019/790; Japan Agency for Cultural Affairs guidance 2023.

You May Also Like

• What Is Artificial Intelligence
• Best AI Tools for Beginners
• How to Use AI
• AI Tools Directory
• Best Free AI Courses

Sources

This article draws on official documentation, product pages, and industry reporting. Specific sources are linked inline throughout the text.

Last reviewed: April 2026