AI Art and Copyright in 2026: Where the Law Stands After the Landmark Rulings

The question that has haunted AI art since DALL-E 2 first went viral — who owns an AI-generated image? — finally has some answers. After years of uncertainty, three major legal developments in late 2025 and early 2026 have established a framework that, while still evolving, gives creators, platforms, and courts clearer guidance than ever before.

The US Copyright Office position

In March 2026, the US Copyright Office issued its most comprehensive guidance to date on AI-generated works. The core principle remains unchanged from its earlier rulings: works created entirely by AI, without sufficient human creative input, are not eligible for copyright protection. But the guidance introduced a crucial nuance — the concept of “meaningful creative control.”

Under the new framework, an AI-generated image can earn copyright protection if a human exercised meaningful creative control over the output. What counts as “meaningful” depends on a multi-factor test: the specificity of the prompt, the number of iterations and selections made, whether the output was further modified by human hand, and whether the AI was used as a tool (like a camera or brush) or as a replacement for human creativity.

The practical effect: an image generated from a single prompt like “sunset over mountains in the style of Ansel Adams” likely won’t qualify. But an image produced through dozens of iterations, with carefully crafted prompts, inpainting, outpainting, manual touch-ups, and selective compositing of multiple AI outputs probably will. The Copyright Office has made clear it will evaluate these on a case-by-case basis, but the framework provides a defensible path to protection for serious AI artists.

The EU AI Act and transparency

The EU’s approach has been different: rather than focusing on copyright eligibility, the AI Act — which came into full effect in early 2026 — mandates transparency. Any AI-generated content made publicly available must be labeled as such. For visual art, this means platforms hosting user-generated content (Instagram, ArtStation, DeviantArt) are required to implement detection and labeling mechanisms.

The transparency requirement has had an interesting side effect: it’s normalizing AI art rather than stigmatizing it. When everything is labeled, the label stops being a scarlet letter and becomes metadata. Several prominent galleries have begun including “AI-assisted” as a medium designation alongside “oil on canvas” and “digital photography,” treating AI tools as just another artistic instrument.

The training data question

The elephant in the room — whether training AI models on copyrighted images constitutes infringement — remains unresolved in most jurisdictions. Several high-profile lawsuits continue to work their way through courts.

The most significant is Getty Images v. Stability AI, which reached a partial settlement in early 2026. Stability AI agreed to license Getty’s catalog for future training and to implement a robust opt-out mechanism for rights holders, but the core legal question — whether the initial unlicensed training was fair use — was left unresolved. The settlement’s structure suggests both parties saw more risk in a definitive ruling than in compromise.

Japan has taken the most permissive stance, explicitly allowing AI training on copyrighted works for non-expressive purposes. The UK is considering a similar approach with an opt-out mechanism. The EU AI Act requires transparency about training data but stops short of requiring licenses for all training data.

What artists should do

For working artists and studios, the practical recommendations are clearer than the law:

First, document your process. Save your prompt iterations, your selection decisions, your manual edits. If you ever need to prove meaningful creative control, the paper trail matters more than the final image.

Second, understand your platform’s terms. Adobe Firefly trains exclusively on licensed and public domain content, making it the safest choice for commercial work. Midjourney and DALL-E operate in more ambiguous territory. If you’re creating for clients with strict IP requirements, the tool you choose matters.

Third, register your copyrights. Even if the registration is initially rejected due to AI involvement, the record of your claim — and the detailed description of your creative process — can become valuable evidence as the law continues to evolve.

The bottom line: AI art isn’t the copyright-free-for-all that early skeptics feared, nor the legal minefield that alarmists predicted. It’s a new medium navigating the same fundamental questions that photography, sampling, and collage faced before it. The law is catching up — slowly, imperfectly, but inexorably.

Primary sources on the U.S. position

• U.S. Copyright Office, “Copyright and Artificial Intelligence, Part 2: Copyrightability” (January 2025): Concludes that copyright protects human authorship; prompts alone generally do not provide sufficient control over outputs, while human-authored inputs, creative selection and arrangement, and modifications of AI output can be protected (copyright.gov/ai).
• Registration guidance (March 2023): Applicants must disclose AI-generated material that is more than de minimis and claim only the human-authored portions (Federal Register).
• Zarya of the Dawn (2023): The Office registered the comic’s human-written text and arrangement but not the individual Midjourney-generated images.
• Thaler v. Perlmutter (D.C. Circuit, March 2025): The appeals court affirmed that a work listing an AI system as its sole author cannot be registered, because the Copyright Act requires human authorship.

The training-data question

Whether training on copyrighted works is fair use is being litigated separately. Cases include Andersen v. Stability AI (artists versus several image-model companies), Getty Images v. Stability AI in the U.S. and the U.K., and the studio suits against Midjourney filed in 2025. In June 2025, a federal judge in Bartz v. Anthropic found training on lawfully acquired books to be fair use but allowed claims over pirated copies to proceed; the case later settled. These decisions are fact-specific and do not settle the question for image models.

Outside the United States

• EU: The AI Act requires providers of general-purpose AI models to publish a summary of training content and to have a policy to respect copyright, including text-and-data-mining opt-outs under the EU copyright directive.
• UK: The Copyright, Designs and Patents Act 1988 contains a provision for “computer-generated works,” a rule that has been under government review.
• China: In 2023, the Beijing Internet Court recognized copyright in an AI-assisted image where the user had made substantial creative choices in prompting and parameter selection.

Practical steps for artists and clients

Keep records of your process (sketches, prompts, edits), disclose AI use when registering works, check each tool’s terms on output ownership and commercial use, and avoid prompts that target living artists’ names or protected characters.