In February 2023, the U.S. Copyright Office issued a landmark ruling on Zarya of the Dawn — a graphic novel submitted by Kristina Kashtanova. The Office initially granted full registration, then rescinded protection for the AI-generated images created with Midjourney, while retaining copyright on Kashtanova's written text and the creative selection and arrangement of panels. The ruling was explicit: because the images were produced by Midjourney's autonomous system, Kashtanova lacked the human authorship required by U.S. copyright law.
That same month, the Copyright Office released formal guidance stating that AI-generated content alone is not copyrightable under current U.S. law, because copyright requires human creative expression. The standard is not whether a human was involved, but whether a human exercised creative control over the final expressive elements.
U.S. copyright law, rooted in the 1976 Copyright Act and interpreted through decades of case law, has always required a human author. The 1884 Supreme Court decision in Burrow-Giles Lithographic Co. v. Sarony established that photographs could be copyrighted because a human made creative choices about lighting, pose, and composition. That principle has never changed — it just now applies to AI tools.
The Copyright Office's March 2023 guidance on AI-generated works distinguishes three categories: (1) purely AI-generated content with no human creative input — not copyrightable; (2) AI-generated content where a human made sufficiently creative choices in selection, arrangement, or modification — potentially copyrightable to the extent of those human choices; (3) human-created work that used AI as a tool, similar to Photoshop filters — likely copyrightable in full.
The key phrase in the guidance is "sufficient human creative control." Typing a short prompt and accepting the AI's first output likely does not meet that standard. Iterating through hundreds of generations, selecting specific outputs, combining them with original writing, and making compositional choices pushes toward protectable expression.
In Thaler v. Vidal (Fed. Cir. 2022), the Federal Circuit affirmed that AI systems cannot be listed as inventors on U.S. patents. Stephen Thaler attempted to list his DABUS AI as the sole inventor of two patents. The court held that "inventors" must be natural persons under the Patent Act. The same human-authorship principle extends to copyright in creative works.
If you are a writer, illustrator, musician, or designer using AI tools in your workflow, the practical implication is this: document your creative choices. Keep records of your prompts, the iterations you rejected, the modifications you made manually, and the editorial decisions that shaped the final work. This documentation becomes your evidence of human authorship if ownership is ever challenged.
For written text, the situation is considerably clearer. If you write text and use AI to polish grammar or suggest synonym choices — and you make the final decisions — the work is yours. If you paste a prompt into an AI and publish the unedited output verbatim, you likely hold no copyright in that output. Between those extremes lies a spectrum of creative collaboration that courts have not yet fully mapped.
Register copyright on your AI-assisted creative work promptly, describe your human creative contributions clearly in the application, and keep a working log of your iterative process. The Copyright Office accepts applications for works containing AI-generated elements — you simply must disclose those elements and claim only the human-authored portions.
You're a creative professional deciding how to protect your AI-assisted work. Describe a specific creative project — a novel, an illustrated article, a music album, a design portfolio — and explain how you used AI in its creation. The assistant will help you analyze which elements are likely copyrightable, which aren't, and how to document your human creative contributions.
In January 2023, three visual artists — Sarah Andersen, Kelly McKernan, and Karla Ortiz — filed suit in the Northern District of California against Stability AI, Midjourney, and DeviantArt, alleging that their artwork was scraped from the internet and used to train image-generation models without consent or compensation. The complaint invoked copyright infringement and right of publicity violations, arguing that the models effectively encoded their artistic styles in a way that allowed users to generate images "in the style of" named artists.
That same month, Getty Images filed suit against Stability AI in the UK High Court of Justice and separately in the U.S. District of Delaware, alleging that Stability AI scraped 12 million images from Getty's licensed library without permission. Getty's complaint included evidence of Stability AI's DALL-E competitor producing images that still contained distorted Getty watermarks — a striking indication of direct training on protected material.
These lawsuits are testing several distinct legal theories simultaneously. The first is direct copyright infringement during training: the argument that copying millions of images or text passages into training datasets, even temporarily, constitutes reproduction under 17 U.S.C. § 106. AI companies counter that this constitutes fair use — transformative use for research and technological development.
The second theory concerns outputs that are substantially similar to training data. This is harder to prove: AI models do not store images literally, but plaintiffs argue that when a model reliably produces outputs "in the style of" a specific artist, the model has encoded protected expression in a form that infringes upon generating similar outputs. Courts have not yet resolved whether this theory survives.
In November 2023, a federal judge in the Andersen v. Stability AI case dismissed most claims but allowed a direct infringement claim by artist Andersen to proceed regarding images that were allegedly reproduced too closely to her specific work. This was a partial victory for plaintiffs — establishing that the door to infringement claims is not closed.
In September 2023, the Authors Guild filed suit against OpenAI, alleging that GPT models were trained on books scraped without authorization. Separately, authors including John Grisham, Jodi Picoult, and George R.R. Martin filed a class action against OpenAI in the same month. These cases focus on text: the argument that a language model trained on complete novels without licensing them infringes the reproduction right, regardless of whether outputs directly copy those novels.
AI companies have consistently advanced fair use as their primary defense. Fair use analysis under 17 U.S.C. § 107 weighs four factors: (1) the purpose and character of the use (commercial vs. educational; transformative vs. reproductive); (2) the nature of the copyrighted work; (3) the amount and substantiality of the portion used; and (4) the effect on the market for the original work.
The strongest argument for AI companies is transformativeness: in Authors Guild v. Google (2d Cir. 2015), the Second Circuit held that Google's mass digitization of books to create a searchable index was transformative fair use. AI companies argue their training use is analogous. Critics counter that unlike Google's index, AI models generate commercial products that directly compete with the original creators' markets — which weighs heavily against fair use on factor four.
No appellate court has yet ruled on AI training fair use. The outcomes of the pending cases — expected to produce significant rulings by 2025 and 2026 — will define the legal landscape for a generation of creative workers.
Register your copyright promptly — registration is a prerequisite for statutory damages and attorney's fees in U.S. infringement suits. Opt out of AI training datasets where opt-out mechanisms exist (Adobe Firefly, OpenAI's web crawler opt-out via robots.txt). Track whether your work appears in training datasets using tools like Have I Been Trained (haveibeentrained.com). Document your publication dates and creative process for any future infringement claims.
You're a professional creative — illustrator, author, photographer, or musician — who has published work online and is concerned about AI training data use. Use this lab to investigate your legal options, understand the current lawsuit landscape, and identify practical protective steps specific to your medium.
In June 2023, the science fiction magazine Clarkesworld — one of the most respected short fiction publications in the field — was forced to temporarily close submissions after receiving a flood of AI-generated stories submitted as human work. Editor Neil Clarke reported that submission volume had increased fivefold, with the vast majority being machine-generated text submitted without disclosure. The magazine had to implement new screening protocols and explicitly ban AI-generated submissions.
In July 2023, Sports Illustrated was found to have published articles under fictional AI-generated author bylines — complete with fabricated headshots produced by an AI avatar service. The publication's parent company, The Arena Group, initially denied the articles were AI-generated before later acknowledging the use of AI content tools. The incident resulted in significant reputational damage and raised serious questions about disclosure obligations in journalism.
The response from professional organizations has been swift. The Authors Guild issued AI disclosure guidelines in 2023, stating that authors should disclose significant AI assistance in the creation of work submitted for publication. The Society of Professional Journalists updated its ethics code guidance to address AI tools, emphasizing that journalists must disclose AI use that materially affects the content of published reporting.
In academic publishing, major journals moved quickly. Nature announced in January 2023 that AI tools cannot be listed as authors, that their use must be disclosed in methods sections, and that authors bear full responsibility for AI-assisted content. Science, The Lancet, and most major academic publishers followed with similar policies within weeks. The International Association of Scientific, Technical and Medical Publishers (STM) issued joint guidance endorsing disclosure as the baseline standard.
For visual artists and designers, the situation is more complex. The Graphic Artists Guild and the American Institute of Graphic Arts (AIGA) have both issued position statements encouraging disclosure when AI generation was used in client work. But no binding professional standard yet governs client contracts in most jurisdictions.
In September 2022, Getty Images banned AI-generated images from its platform entirely, citing legal uncertainty and the rights of contributing photographers. In 2023, Getty reversed course and launched its own licensed AI image generator trained only on its licensed content, with compensation to contributing photographers built into the model. This illustrates how platforms are being forced to develop explicit disclosure and compensation frameworks — creating a precedent for the broader industry.
For freelancers and agencies, disclosure to clients is increasingly a contractual and ethical obligation. Several major advertising agencies — including WPP and Publicis — have developed internal AI use policies that require disclosure to clients when AI tools are used in campaign creation. The underlying concern is twofold: clients want to ensure they are receiving what they are paying for, and clients need to know about potential copyright complications in AI-generated elements they intend to use commercially.
A practical standard emerging in commercial creative work is the "material assistance" threshold: if AI performed work that a human would otherwise have been paid to perform (writing copy, generating images, composing music), that use should be disclosed. If AI was used as a productivity tool analogous to spell-check or a thesaurus, disclosure is generally not required by emerging norms — though some clients contractually require disclosure of any AI use.
Check your contracts carefully. An increasing number of publishers, studios, and clients are including explicit AI disclosure and warranty clauses, requiring creators to warrant that delivered work was not substantially created by AI without prior written approval.
When in doubt, disclose upward — to clients, editors, and audiences — rather than downward. Describe specifically what AI did (generated image drafts, suggested structural outlines, produced alternative copy options) and what you did (selected, modified, directed, edited, approved). This framing positions AI as a tool you deployed rather than a replacement for your creative judgment — which is both more honest and more protective of your professional reputation.
You need to communicate your AI use to a client, editor, publisher, or audience. This lab helps you craft clear, professional disclosure language appropriate to your specific situation — one that is honest, specific about what AI did and what you did, and positions your creative judgment as the driving force.
In July and August 2023, the Screen Actors Guild–American Federation of Television and Radio Artists (SAG-AFTRA) struck against the major studios in part over AI protections — specifically, the right of studios to scan an actor's likeness and use it in AI-generated performances without additional compensation or consent. The strike, which ended in November 2023 with a new contract, won explicit protections: studios must obtain consent before creating a digital replica of a performer's likeness, and residuals must be paid when such replicas are used. The agreement established a precedent for consent and compensation frameworks across the entertainment industry.
Simultaneously, the Writers Guild of America (WGA), which had been on strike since May 2023, secured in its September 2023 contract that AI cannot write or rewrite literary material used in Guild-covered productions, that AI-generated material does not constitute "literary material" under the MBA (Minimum Basic Agreement), and that writers cannot be required to use AI tools. The WGA deal also required studios to disclose when AI-generated material is given to writers as source material.
The SAG-AFTRA and WGA agreements set a template that is influencing contracts far beyond Hollywood. The provisions that matter most for working creatives fall into several categories:
AI Generation Restrictions: Language stating that AI cannot be used to generate, substantially rewrite, or replace the contracted human's creative output without written approval. This is now standard in many book publishing contracts, particularly for major commercial publishers, and is being negotiated into music recording agreements.
Likeness and Voice Protections: Post-SAG-AFTRA, contracts in music, voice acting, and modeling increasingly include explicit prohibitions on creating AI replicas of an individual's likeness, voice, or performance style without consent and additional compensation.
Training Data Exclusions: Contract language specifying that delivered work product cannot be used to train AI models — by the contracting party, its licensees, or any third parties. This has become a standard ask in contracts between creators and tech platforms.
IP Ownership Clarity: Language explicitly establishing that the creator — not the AI tool used — owns the copyright in the delivered work, and that the contracting party receives only the rights explicitly granted.
Every AI tool's Terms of Service contains provisions about ownership of outputs and potential use of your inputs. OpenAI's Terms (as of 2024) state that users own their outputs and OpenAI does not claim rights in them, but that OpenAI may use inputs to improve services. Adobe Firefly's Terms state that images generated using licensed content are safe for commercial use and indemnified by Adobe. Midjourney's Terms previously granted users only a limited license to outputs (with full commercial use requiring a paid plan) and included broad rights for Midjourney to use generated images. Read the ToS of every tool you use commercially. These terms change — set a calendar reminder to review them annually.
If you are licensing — rather than selling outright — your creative work, the license agreement must address AI specifically. A robust license for AI-assisted creative work should: (1) specify which elements of the work are human-authored and which were AI-generated; (2) limit the licensee's right to use the work as training data for AI systems; (3) include a "no derivative AI works" clause if you do not want the licensee to feed your work into an AI to generate variations; and (4) require the licensee to pass the training-data restriction through to any sublicensees.
For musicians and songwriters, the emergence of AI voice-cloning and style-transfer tools has made it essential to address these explicitly in sync licenses, recording contracts, and producer agreements. A growing number of music publishers are including "AI voice clone" restrictions in standard deal language — prohibiting licensees from creating AI-generated performances in an artist's voice using licensed sound recordings as training material.
Register copyrights promptly with the Copyright Office, disclosing AI elements and claiming human-authored portions. Document your creative process — prompts, iterations, decisions. Implement robots.txt opt-outs. Review every AI tool's Terms of Service annually. Add AI-specific provisions to your contracts: training data exclusions, digital replica protections, IP ownership clarity. Disclose AI use to clients and editors at the "material assistance" threshold. Follow the SAG-AFTRA and WGA precedents: consent, disclosure, and compensation are the three pillars of fair AI use in professional creative work.
You need to add AI-protective provisions to a contract you're negotiating — or reviewing one that a client has sent you. This lab helps you draft specific clause language for training data exclusions, digital replica protections, IP ownership clarity, and disclosure obligations appropriate to your creative field.