In February 2023, the US Copyright Office issued a landmark ruling on Zarya of the Dawn, a comic book written by Kristina Kashtanova using text she wrote and images generated by Midjourney. The Office granted copyright to Kashtanova's written text and her arrangement of images — but explicitly refused copyright for the AI-generated images themselves, because they were not the product of human authorship. This was the first formal US ruling applying copyright doctrine to AI-generated visual art.
The decision sent a clear signal across the creative industry: prompts alone do not make you an author under current US law. The creative expression must come from a human being, not from a model predicting likely pixels.
US copyright law has a foundational requirement that traces to the Supreme Court's 1884 ruling in Burrow-Giles Lithographic Co. v. Sarony: authorship means human creative expression. The Copyright Office's 1973 compendium — updated repeatedly since — explicitly states that copyright cannot protect works produced by machines without creative input from a human author.
When you type a prompt into an image generator, the model selects outputs based on statistical patterns learned from training data. You describe; the model decides. Courts and the Copyright Office have consistently held that this description-to-execution chain does not qualify as the kind of "original creative expression" that copyright was designed to protect.
This matters enormously for anyone building products, publishing work, or licensing content created with AI assistance. Work without copyright protection can be freely copied by anyone.
Stephen Thaler attempted to register copyright for an image produced entirely by his DABUS AI system, listing the AI as author. The DC Circuit Court affirmed the Copyright Office's refusal: "Human authorship is a bedrock requirement of copyright." The case is ongoing in various forms but the principle has held across every US federal ruling to date.
Copyright law does not treat all AI involvement the same way. Think of it as a spectrum:
You write every word; use AI only for grammar checks. Your copyright is clear. The AI is acting as a spellchecker, not a creative contributor.
AI generates many options; you curate, edit, arrange. Copyright may cover your selection and arrangement — as in Kashtanova's comic layout.
You write a prompt; AI produces the final work unchanged. No copyright for the AI output under current US Copyright Office guidance.
No human prompt or direction at all. No copyright exists. The work enters the public domain immediately upon creation.
If you are making something real — a blog, a game, a product — you need to think about the copyright status of every asset. AI-generated images used in a commercial product you intend to protect can be freely copied by competitors unless you have added sufficient human creative expression on top of them.
Practical strategies that strengthen your copyright claim: write your own captions, arrange AI-generated elements in original compositions, substantially edit and modify AI drafts, combine AI output with original photography or illustration, and document your creative decisions along the way.
Your creative choices are the asset. The more your final work reflects decisions only you could have made — your taste, your judgment, your editing hand — the stronger your claim to ownership. Use AI as a powerful tool, but keep yourself firmly in the creative driver's seat.
Think of a real or hypothetical creative project you might build with AI — a blog, a game, a product listing, a piece of art, a story. Describe it to the AI advisor below and explore where your copyright protection is strong and where it may be thin.
Have at least 3 exchanges. The advisor will help you identify which parts of your project are human-authored and which might be considered AI-generated under current US Copyright Office standards.
In January 2023, three artists — Sarah Andersen, Kelly McKernan, and Karla Ortiz — filed a class action lawsuit in the Northern District of California against Stability AI, Midjourney, and DeviantArt, alleging that these companies had scraped billions of images from the internet without consent to train their models. The artists argued this constituted copyright infringement at massive scale — that each training image was copied into model weights without a license.
The same month, Getty Images filed a separate suit against Stability AI in the UK High Court and the US District of Delaware, noting that Stability AI's model could generate images bearing distorted Getty watermarks — suggesting the training data included watermarked Getty photos that were copied directly. Getty alleged infringement of over 12 million photographs.
These cases are testing three distinct legal theories simultaneously, and courts are still working out which ones hold:
Copying images into training datasets without a license. If each image is a copy, each copy may be infringement. Scale: Stable Diffusion's dataset contained ~5.8 billion image-text pairs.
AI outputs that are substantially similar to copyrighted training works. This requires proving the model memorized and reproduced — not just learned style from — a specific work.
Attempting to protect artistic style, not just specific works. Courts have historically held that style alone cannot be copyrighted — only specific expression in a particular work.
AI companies have relied heavily on the fair use doctrine, arguing that training on copyrighted works is transformative — similar to how Google was allowed to scan books for search indexing (Authors Guild v. Google, 2015). Under fair use, courts weigh four factors: the purpose and character of the use, the nature of the copyrighted work, the amount copied, and the effect on the market for the original.
The transformative use argument received a blow in 2023 when the Supreme Court ruled in Andy Warhol Foundation v. Goldsmith that commercial use is not automatically transformative. This ruling, though not directly about AI, narrowed the fair use defense that AI companies had been counting on. Legal scholars are divided on how it will affect pending AI training cases.
The New York Times sued OpenAI and Microsoft, alleging that GPT-4 was trained on millions of Times articles without permission and could reproduce them nearly verbatim. The complaint included exhibits showing ChatGPT producing extended passages matching Times articles word-for-word — strengthening the "memorization" argument that goes beyond mere style learning. This case is widely considered the highest-stakes AI copyright case currently active.
As a creator using AI tools, you sit in an interesting middle position. Your work may have been used to train the models you're now using — without your consent or compensation. At the same time, the outputs you produce may face claims if they too closely replicate specific copyrighted works.
Some AI companies have introduced opt-out registries (Adobe's Content Credentials, Spawning's "Have I Been Trained" database) that allow artists to request their work be excluded from future training. These are voluntary and not legally binding — but they represent an emerging norm.
When you use AI for commercial work, reviewing the model provider's terms of service matters. Many tools now explicitly address ownership and indemnification in their commercial tiers — Midjourney's paid tier grants commercial rights; their free tier does not.
None of the major AI copyright cases have reached final verdicts as of this writing. The legal landscape is shifting quickly. What is certain: these cases will define the rules of AI-assisted creation for a generation. Stay current with outcomes — they will directly affect what tools you can use commercially and on what terms.
You're going to have a conversation about the training data side of AI copyright. Pick an AI tool you use or are curious about — an image generator, a writing assistant, a music AI — and explore the legal and ethical dimensions of what it learned from.
Have at least 3 exchanges. Ask about specific tools, specific lawsuits, or what due diligence you should do before using AI-generated content commercially.
In June 2023, the science fiction magazine Clarkesworld temporarily closed submissions after being overwhelmed by a flood of AI-generated stories — editor Neil Clarke reported receiving hundreds of AI submissions per day, a volume that made human review impossible. He identified the submissions not because they were labeled as AI-generated, but because their writing patterns were recognizable.
Around the same time, the Associated Press, the BBC, and other major outlets began publishing explicit AI-use policies for their journalists. The AP's policy, released in August 2023, permitted AI assistance for research but prohibited publishing AI-generated text as original reporting and required disclosure when AI tools substantially shaped a piece. The policy stated clearly: "the AP's reputation rests on the trust audiences place in the humans who report and edit our work."
As of this writing, there is no US federal law requiring creators to disclose when content was AI-generated — with one significant exception: the Federal Trade Commission's guidelines on endorsements and testimonials require disclosure when AI is used in advertising in ways that could be materially deceptive. Several US states have introduced AI disclosure bills; none has become comprehensive law.
The European Union's AI Act, which passed in 2024, requires that AI-generated content be labeled as such when it could be confused with genuine human content — particularly for deepfakes and synthetic media in political contexts. This is binding law for any creator distributing content in EU markets.
Synthetic media — AI-generated video or audio that realistically depicts a real person doing or saying something they did not do — faces stricter and faster-developing regulation. As of 2024, over 20 US states have laws specifically addressing deepfakes in political advertising and non-consensual intimate imagery. Several platforms (YouTube, Meta, TikTok) now require creators to label AI-generated realistic content under their community standards.
For most working creators, platform policies matter more immediately than legislation. The major platforms have moved faster than legislators:
Requires creators to disclose AI-generated realistic content — faces, voices, real events. Failure to disclose can result in removal and account suspension.
Requires labels on AI-generated video, audio, and images that are "photorealistic." Uses both manual disclosure and automated detection.
Requires disclosure of AI-generated content in books listed on Kindle Direct Publishing. Does not prohibit AI content but mandates flagging it.
Getty, Shutterstock, and Adobe Stock each have distinct policies — some accept labeled AI images; others ban them entirely from certain categories. Check per-agency rules.
Legal compliance is the floor, not the ceiling. The creative community is developing its own norms faster than law can codify them. Most working professional communities — journalism, academic publishing, illustration, music — have arrived at a common default: disclose meaningfully when AI substantially contributed to what your audience believes came from you.
The practical test many creators apply: Would my audience feel deceived if they found out AI generated this and I didn't tell them? If yes, disclose. This is not about penalizing AI use — it is about maintaining the trust relationship with audiences that gives creative work its value in the first place.
Attribution also has a positive framing: crediting your workflow honestly can be a competitive signal. Increasingly, audiences and clients want to know what they are paying for. A portfolio that clearly shows what is human work and what is AI-assisted is more legible and more trustworthy than one that obscures the distinction.
Check three things before publishing AI-assisted content: (1) the platform's current AI policy, (2) any applicable law in your distribution territory, and (3) the professional norms of your field. When in doubt, disclose. Disclosure rarely costs you much; undisclosed AI work discovered after the fact can cost you everything.
You're going to draft an AI disclosure statement for a real or hypothetical creative project — a blog, a product, a social media account, a publication. The AI advisor will help you figure out what to say, how to say it, and whether it matches the platform and legal requirements that apply to your situation.
Have at least 3 exchanges. Push back, ask for examples, and refine your draft through the conversation.
In 2023, musician Grimes made an unusual announcement: she would allow anyone to use her AI-cloned voice to create songs, provided they split streaming royalties 50/50 with her. She released an AI voice model trained on her vocals and invited the public to generate music with it. Within months, thousands of AI-Grimes songs had been created and uploaded. The experiment was part licensing innovation, part career strategy — positioning Grimes as an artist whose brand transcended the constraint of being the only one who could produce her sound.
At the same time, Universal Music Group sent cease-and-desist demands to AI music platforms and pressured streaming services to remove AI-generated tracks that mimicked the voices of signed artists — without those artists' consent. The contrast between Grimes (opt-in) and UMG's roster (opt-out) illustrated the two poles of the emerging debate: controlled licensing vs. defensive protection.
Despite the legal uncertainty around AI-generated content, your human creative work remains fully protectable. Strategic registration with the US Copyright Office costs $65–$85 per application and creates a public record, enables you to sue for statutory damages (up to $150,000 per infringement for willful violations), and establishes the date of your authorship. For commercial creative work, registration is worth the cost.
Trademark protection is complementary and in some ways more powerful for AI era concerns: your name, brand, logo, and distinctive marks can be trademarked. Trademarks protect identity in commerce — and AI voice or style imitation that creates consumer confusion in commercial contexts may constitute trademark infringement independent of copyright.
Separate from copyright and trademark, the right of publicity protects your identity — name, likeness, voice, and persona — from commercial exploitation without consent. It's a state-law right in the US (California and New York have the strongest protections) and has become newly important in the AI era. In 2023, voice actors, musicians, and celebrities began asserting right of publicity claims against AI voice cloning tools that recreated their voices for commercial use without permission.
A practical ecosystem of tools is emerging to help creators protect their work from AI training and assert provenance of their human-made content:
A searchable database of training images. Artists can opt out of future Stable Diffusion training runs by registering their work. Voluntary, not legally enforceable — but widely respected.
A metadata standard attaching creator information, editing history, and AI-use disclosure to image files. Implements the C2PA open standard — increasingly adopted by cameras, phones, and platforms.
Tools from the University of Chicago that add invisible perturbations to images, disrupting how AI models learn from them if scraped. Actively degrades AI training on your specific work.
Standard website configuration now includes options to block AI training crawlers. OpenAI, Common Crawl, and others respect these. Most CMS platforms now offer AI-crawl blocking as a standard setting.
No single tool or legal doctrine provides complete protection. A layered approach makes sense: register your most valuable original works with the Copyright Office; trademark your name and brand; use content credentials on commercial images; deploy technical protections like robots.txt blocking on your website; and consider opt-out registries for any training dataset you can identify.
Most importantly: keep making work that is distinctively, demonstrably yours. The creative choices, personal perspective, and human judgment you bring to your work are not just copyright strategy — they are the actual reason audiences choose your work over generically AI-generated alternatives. In an era of infinite AI-generated content, distinctively human work becomes more valuable, not less.
Grimes's experiment showed that in the AI era, the most powerful form of protection may not be legal defense but strategic positioning. If you control the terms on which AI engages with your creative identity — rather than only reacting when it happens without you — you retain agency over your own brand. Protection and participation are not opposites. The key is that the choice is yours to make.
You're going to build a real protection plan for your creative work. This covers copyright registration priorities, trademark considerations, technical tools, and how you want to position yourself relative to AI use of your work.
Have at least 3 exchanges. Tell the advisor what you make, what matters most to protect, and what your resources and constraints are — they'll help you build a tiered, practical plan.