Sophia HamiltonUnder current global copyright law, AI cannot own copyright — only humans and legal entities can. As of 2026, content generated entirely by AI without meaningful human creative input is either unprotected or falls into the public domain the moment it is created. But the law is catching up fast: courts in the United States, the European Union, and the United Kingdom have issued landmark rulings in the past 18 months that are redefining what counts as "human authorship." For creators, legal teams, and businesses that rely on AI tools, understanding where intellectual property rights begin and end is no longer optional.
The Human Authorship Problem at the Core of Copyright Law
Copyright law across every major jurisdiction rests on a single foundational assumption: that creative works originate from a human mind. In the US, Section 102(a) of the Copyright Act of 1976 protects "original works of authorship fixed in any tangible medium." The US Copyright Office (USCO) has interpreted this consistently to require human authorship — in 2023, it refused to register an AI-generated comic strip's images while granting protection only to the human-authored text surrounding them.
The same logic governs European Union law under Directive 2001/29/EC, which requires that a work reflect the author's "own intellectual creation." The Court of Justice of the European Union (CJEU) has repeatedly held that "intellectual creation" demands a human being expressing their personality through free and creative choices. An algorithm optimizing for a statistical output does not satisfy this test.
What makes 2026 different is not a change in the core standard — it is the volume and commercial significance of AI-generated content now reaching courts and registries. Over 40% of digital content published by US media companies in 2025 involved at least one AI-generation step, according to a Reuters Institute survey. That volume has forced regulators to move from broad policy statements to line-by-line adjudication.
The critical legal question has sharpened into a precise inquiry: how much human creative input is enough to give a human (or company) ownership of work that an AI substantially produced?
What US, EU, and UK Copyright Law Actually Says in 2026
United States: The USCO's Evolving Threshold
The US Copyright Office finalized its guidance on AI and copyright in February 2025, confirming that purely AI-generated content cannot be registered. However, it established a spectrum: works where a human exercised "sufficient creative control" — selecting, arranging, modifying, or adding to AI output — remain eligible for protection. The human's protectable contribution is the only thing covered; the AI-generated elements themselves enter the public domain.
Courts are now wrestling with what "sufficient" means. In Thaler v. Perlmutter (D.D.C. 2025 appeal pending), the court affirmed that an image produced entirely by an AI system without human creative choices was not copyrightable. By contrast, a separate USCO ruling in late 2024 granted registration to a graphic novel where the author used AI to generate initial image layers, then manually altered composition, color grading, and narrative framing — ruling that the human's modifications met the threshold.
European Union: The "Own Intellectual Creation" Standard
The EU's approach under the EU AI Act (which entered full application in 2026) does not directly address copyright ownership but introduces transparency obligations: providers of general-purpose AI systems must publish summaries of training data used. This intersects with copyright because training data may include protected works. The EU is separately advancing a proposal to create a narrow sui generis right for "AI-assisted works," mirroring the existing database right under Directive 96/9/EC.
United Kingdom: The Computer-Generated Works Provision
The UK's Section 9(3) of the Copyright, Designs and Patents Act (CDPA) 1988 already provides that for computer-generated works, copyright belongs to "the person by whom the arrangements necessary for the creation of the work are undertaken." This makes the UK the most permissive major jurisdiction for AI content ownership — the developer or operator of the AI can potentially claim copyright even without traditional human authorship. However, the UK Intellectual Property Office (IPO) published a consultation in 2025 questioning whether this provision remains fit for purpose in the generative AI era.
Who Can Actually Own AI-Generated Content: Three Contenders
When AI produces a piece of writing, an image, or a musical composition, three parties commonly argue for ownership. Each claim has legal merit in some contexts and fails in others.
The AI developer — companies like OpenAI, Google DeepMind, or Stability AI — typically hold the strongest contractual position through terms of service. Most major AI platforms state that output belongs to the user, but the developer retains a broad license. Where a developer builds an AI system specifically to produce commercial outputs (stock images, code, reports), they may assert the UK-style "arrangements" claim.
The user or prompter is the party most courts and regulators are now focusing on. A detailed, iterative prompt that substantially directs the AI's creative choices — specifying style, narrative arc, color palette, mood — creates a record of human creative input. The more specific and consequential the human direction, the stronger the claim. A single-sentence prompt generating a generic landscape has little chance of copyright protection; a 500-word prompt specifying exact visual elements, followed by multiple refinement cycles, has a credible claim in US and UK law.
Nobody — public domain is the default outcome under US law when human input is minimal. This matters commercially: if a company uses AI to mass-produce marketing copy without significant human editing, competitors can legally reproduce that content. Businesses that do not understand this risk losing a key competitive asset the moment they publish.
Expert perspective: "The critical mistake companies make is assuming AI output is automatically theirs. In the US, if your team gives the AI a vague instruction and publishes the result unchanged, you own nothing. Your legal team needs to document every creative decision a human made — that documentation is your copyright claim." — IP attorney specializing in technology and media law, 2026
À retenir: Human authorship documentation is not just a legal formality — it is the only mechanism by which AI-generated work gains copyright protection under US law.
The Global Harmonization Challenge: WIPO and Beyond
No international treaty currently governs AI and intellectual property rights, creating a patchwork of national approaches that complicates cross-border commercial use of AI-generated content. The World Intellectual Property Organization (WIPO) launched its formal consultation on "Generative AI and Intellectual Property" in 2024, with member states submitting position papers through early 2026. The outcome is expected to be a non-binding recommendation rather than a treaty amendment — meaningful progress, but not enforceable uniformity.
Consider a concrete scenario: a UK-based fintech company uses a US AI platform to generate investor communications, then distributes them to EU clients. Under UK law, the company may hold copyright through the CDPA arrangement provision. Under US law, the communications are in the public domain. Under emerging EU proposals, the AI provider must disclose training data — which may reveal third-party copyrighted material used to train the model, creating an infringement risk at the training layer rather than the output layer.
This jurisdictional fragmentation is not academic. In 2025, the EU's Data Act came into force requiring data-generated content to be traceable, which AI providers must now reconcile with their copyright disclaimers. Companies operating across borders need jurisdiction-specific IP strategies, not a single global policy.
Japan and China, notably, have both moved toward more permissive frameworks. Japan's Cultural Affairs Agency clarified in 2024 that AI-generated content can receive copyright protection when a human's "creative contribution" is evident, without specifying a minimum threshold. China's National Copyright Administration issued draft rules in 2025 proposing that AI-assisted works be protected if a human directed the AI in a "creative manner."
Practical Steps to Protect AI-Assisted Work in 2026
For creators, legal teams, and businesses, the current legal environment rewards documentation and deliberate process. The following steps reflect best practice given the 2026 state of law in the US, EU, and UK.
1. Record the creative process, not just the output. Save every prompt iteration, every refinement instruction, every manual edit applied to AI output. Courts and registries are increasingly looking for a "creative decision trail" that shows a human was directing — not simply accepting — what the AI produced.
2. Apply meaningful human modification before publication. Editing AI output for tone, accuracy, structure, and style — and being able to show what changed — is the most reliable way to establish copyright eligibility under US law. Wholesale publication of unedited AI output forfeits protection.
3. Review your AI provider's terms of service. Most generative AI platforms assign output ownership to the user but retain broad license rights. Read the applicable terms for the specific version and date of the AI tool used — these terms change, and the version you used in 2024 may have had different ownership language than the 2026 version.
4. Consider trade secrets for AI workflows. Even where copyright does not apply to output, the proprietary prompt engineering, fine-tuned models, and production workflows a company develops can be protected as trade secrets under the Defend Trade Secrets Act (DTSA) in the US or equivalent national laws.
5. Obtain IP warranties in B2B AI contracts. When purchasing AI-generated deliverables from a vendor, require contractual representations that the output does not infringe third-party IP and that the vendor has obtained appropriate training data licenses.
Disclaimer: This article provides general legal information only and does not constitute legal advice. AI and intellectual property law varies significantly by jurisdiction and is changing rapidly. Consult a qualified intellectual property attorney for advice specific to your situation and territory.
Frequently Asked Questions About AI Copyright Law in 2026
Can an AI system hold copyright on content it creates?
No. Every major jurisdiction — including the US, EU, UK, Japan, and China — requires human (or legal entity) authorship for copyright protection. AI has no legal personhood and cannot hold IP rights. This position is unanimous globally as of 2026 and is unlikely to change in the near term.
Is all AI-generated content automatically in the public domain?
Not necessarily. In the US, purely AI-generated content with no meaningful human creative input enters the public domain. But content where a human exercised sufficient creative control — directing, selecting, modifying — can be copyrightable. In the UK, the CDPA's computer-generated works provision allows the "arranger" of an AI system to hold copyright even without traditional human authorship.
What counts as "sufficient" human creative input under US copyright law?
The US Copyright Office has not set a precise numerical threshold. What matters is whether a human made expressive choices: selecting which AI outputs to use, arranging them in a specific order, modifying content in ways that reflect personal creative judgment, or combining AI output with original human-authored material. Detailed iterative prompting that steers the AI's creative direction has been treated favorably in recent USCO review decisions.
Who is liable if AI-generated content infringes a third-party copyright?
Liability depends on the relationship between parties. AI developers face potential liability at the training data layer if their model was trained on copyrighted material without license or fair use justification. Users face potential liability if they deploy AI output commercially and the output reproduces protected expression. Courts are still developing the applicable framework — several major cases are currently pending in US federal courts as of mid-2026.
Does the EU AI Act create new intellectual property rights?
No. The EU AI Act (fully applicable from August 2026) focuses on risk classification, transparency, and safety obligations for AI systems — it does not directly create or modify intellectual property rights. However, its transparency requirements (including disclosure of training data summaries) interact with copyright law and may support infringement claims based on unlicensed training data.
