L1
·
Quiz
·
Lab
L2
·
Quiz
·
Lab
L3
·
Quiz
·
Lab
L4
·
Quiz
·
Lab
Module Test
Module 3 · Lesson 1

The Authorship Question

When a machine writes your words, who holds the pen — and does copyright even care?
If you describe a painting and AI renders it, is the result yours?

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.

Why Copyright Requires a Human Author

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.

Key Ruling — Thaler v. Vidal (2022)

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.

The Spectrum: From Tool to Author

Copyright law does not treat all AI involvement the same way. Think of it as a spectrum:

Heavy Human Control

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.

Significant Human Selection

AI generates many options; you curate, edit, arrange. Copyright may cover your selection and arrangement — as in Kashtanova's comic layout.

Prompt-Only Generation

You write a prompt; AI produces the final work unchanged. No copyright for the AI output under current US Copyright Office guidance.

Fully Autonomous AI

No human prompt or direction at all. No copyright exists. The work enters the public domain immediately upon creation.

What This Means When You Create

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.

Authorship threshold
The minimum level of human creative expression required for copyright protection to attach to a work. Currently interpreted as requiring that a human being make the original creative choices.
Public domain
Works not protected by copyright — anyone can use, copy, and build on them freely. Fully AI-generated works in the US currently fall here immediately.
Selection and arrangement
A doctrine allowing copyright in the original way a creator selects, orders, or combines elements — even if individual elements are not copyrightable. Used by the Copyright Office to protect Kashtanova's comic layout while rejecting her AI images.
Bottom Line

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.

Quiz — The Authorship Question

3 questions · select the best answer
In the Zarya of the Dawn ruling, what did the US Copyright Office decide to protect?
Correct. The Office applied the "selection and arrangement" doctrine to protect Kashtanova's human creative choices — her text and how she laid out the pages — while explicitly refusing protection for the AI-generated images, since those were not produced by human authorship.
Not quite. The Office used a nuanced ruling: human-authored elements (text, layout) were protected, but the Midjourney images were not, because they lacked human authorship.
Under current US Copyright Office guidance, which scenario gives you the weakest copyright claim?
Correct. Prompt-only generation with no human editing or arrangement is the scenario the Copyright Office specifically cited when refusing protection: the human is describing, not creating. The output lacks human authorship and enters the public domain.
Think about which scenario leaves the least human creative input in the final work. The Copyright Office's concern is whether a human made the original creative choices that appear in the output.
What was the core principle affirmed in Thaler v. Vidal?
Correct. The DC Circuit used those exact words — "bedrock requirement" — affirming the Copyright Office's refusal to register an AI-authored work. Neither the AI nor its owner could hold copyright simply by owning the machine that produced the work.
The Thaler case is significant precisely because it rejected the idea that AI ownership transfers to copyright. The court's language was unambiguous: human authorship is required.

Lab 1 — Mapping Your Copyright Exposure

Chat with an AI advisor about your creative project and authorship

Your Task

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.

Starter: "I'm planning to create [describe your project]. I'll be using AI to [describe how]. Help me understand which parts I can copyright."
Copyright Advisor
AI Lab
Hello! I'm your copyright advisor for this module. Tell me about a creative project you're planning to build with AI — what it is, what you'll make yourself, and where AI will be doing the heavy lifting. I'll help you map out where your copyright is solid and where it might be thin under current US Copyright Office standards.
Module 3 · Lesson 2

Training Data and the Copying Question

AI models learned from someone's work — and those someone's are now asking courts what that means.
If an AI trained on your art can reproduce your style, has it stolen from you?

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.

The Three Legal Theories in Play

These cases are testing three distinct legal theories simultaneously, and courts are still working out which ones hold:

Direct Infringement

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.

Output Infringement

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.

Style as Protectable Expression

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.

Fair Use and the Transformative Defense

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 v. OpenAI (December 2023)

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.

What This Means for Your Work

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.

Fair use
A legal doctrine permitting limited use of copyrighted material without permission, evaluated across four factors: purpose, nature of work, amount taken, and market effect.
Transformative use
A use that adds new meaning, expression, or message to the original. More transformative uses are more likely to qualify as fair use — but commercial purpose weighs against this.
Memorization
When an AI model reproduces training data nearly verbatim rather than learning general patterns. Memorization strengthens infringement claims because it demonstrates actual copying, not just style learning.
The Law Is Not Settled

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.

Quiz — Training Data and Copying

3 questions · select the best answer
The Getty Images lawsuit against Stability AI was strengthened by evidence that the model could generate images containing what?
Correct. The presence of distorted Getty watermarks in generated images was cited as direct evidence that the training dataset included watermarked Getty photographs — meaning specific protected works were copied, not just learned from abstractly.
The smoking gun in the Getty case was the watermarks. When an AI generates an image with a recognizable but distorted watermark, it suggests the original watermarked image was in the training data and was memorized to some degree.
Why did the Supreme Court's 2023 ruling in Andy Warhol Foundation v. Goldsmith matter for AI copyright cases?
Correct. While the Warhol case was about a Prince portrait, not AI, it weakened the "transformative commercial use" argument that AI companies had been leaning on. Courts can now more readily find that commercial AI training is not automatically protected as fair use.
The Warhol ruling didn't address AI directly, but it affected a core defense AI companies were planning to use. Think about what the ruling said about commercial uses of copyrighted work.
Under long-established copyright doctrine, which of the following is generally not protectable by copyright?
Correct. Courts have consistently held that style, technique, or aesthetic approach cannot be copyrighted — only specific expression in a particular work. This is why artists suing over "style imitation" face an uphill battle, even when AI clearly learned from their distinctive work.
This is a key distinction in copyright law. Specific works can be protected, but broader artistic style — the "look and feel" that doesn't appear verbatim in any one work — has historically been outside copyright's reach.

Lab 2 — Evaluating Training Data Risk

Discuss the copyright implications of AI training data for your use case

Your Task

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.

Starter: "I use [tool name] to generate [type of content] for [commercial/personal] use. What do I need to know about its training data and whether that creates legal risk for me?"
Training Data Advisor
AI Lab
Ready to dig into training data and copyright risk. Tell me which AI tool you're using and what you're creating with it — I'll walk you through the current legal landscape, what the ongoing cases mean for your specific use case, and practical steps to reduce your exposure.
Module 3 · Lesson 3

Attribution, Credit, and the Disclosure Debate

Audiences are asking whether what they're consuming was made by a person — and the answer is getting harder to give.
Do you have an obligation to tell people when AI made what you're showing them?

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."

No Universal Legal Requirement — Yet

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.

The Deepfake Distinction

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.

Platform Policies as De Facto Law

For most working creators, platform policies matter more immediately than legislation. The major platforms have moved faster than legislators:

YouTube (2023)

Requires creators to disclose AI-generated realistic content — faces, voices, real events. Failure to disclose can result in removal and account suspension.

Meta (2024)

Requires labels on AI-generated video, audio, and images that are "photorealistic." Uses both manual disclosure and automated detection.

Amazon KDP (2023)

Requires disclosure of AI-generated content in books listed on Kindle Direct Publishing. Does not prohibit AI content but mandates flagging it.

Stock agencies

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.

The Ethics Beyond the 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.

Synthetic media
AI-generated content that realistically depicts people, events, or environments — including deepfake video, voice cloning, and photorealistic generated imagery.
Material deception
Deception that could affect a reasonable person's decisions or beliefs. The FTC standard for requiring disclosure — if AI use in advertising could materially mislead consumers, it must be disclosed.
Practical Rule

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.

Quiz — Attribution and Disclosure

3 questions · select the best answer
What specific event prompted Clarkesworld to temporarily close submissions in early 2023?
Correct. Editor Neil Clarke reported hundreds of AI-generated submissions per day. The volume made meaningful human editorial review impossible, forcing a temporary closure — a vivid illustration of how AI is already disrupting the creative economy at the submission/gatekeeping stage.
The Clarkesworld case was specifically about submission volume. The magazine wasn't responding to a legal or policy change — it was responding to a practical operational crisis caused by the sheer number of AI-generated stories being submitted.
Under the EU AI Act passed in 2024, which type of content specifically requires disclosure labeling?
Correct. The EU AI Act targets content where the deception risk is highest — synthetic media that realistically mimics humans, especially in political contexts where manipulation of public opinion is a concern. It doesn't catch every AI-assisted creation, but it does impose binding obligations on realistic synthetic content distributed in EU markets.
The EU AI Act is targeted rather than comprehensive. It focuses on the highest-risk category: content that could deceive audiences about its origin, particularly realistic synthetic media used in politically sensitive contexts.
What is the practical ethics test many professional creators apply when deciding whether to disclose AI involvement?
Correct. This audience-trust test — "would they feel deceived?" — captures the ethical core of the disclosure debate better than any percentage threshold or technical definition. It centers the relationship between creator and audience, which is ultimately what professional reputation depends on.
The most useful practical test isn't about percentage or technical detectability — it's about trust. If discovering the AI involvement after the fact would make your audience feel misled, that's the signal to disclose proactively.

Lab 3 — Writing Your Disclosure Policy

Draft a real disclosure statement for your creative work with AI coaching

Your Task

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.

Starter: "I run [describe your creative project or outlet]. I use AI to [describe uses]. Help me write a disclosure statement that's honest, professional, and appropriate for [my audience / platform]."
Disclosure Drafting Assistant
AI Lab
Let's write a disclosure statement that actually works for you. Tell me about your creative project — what you make, how AI is involved, and who your audience is. I'll help you draft something honest, appropriately specific, and calibrated to any platform rules or legal requirements that apply.
Module 3 · Lesson 4

Protecting Your Work in the AI Era

You can't stop AI from learning from the world — but you can make deliberate choices about what you share, register, and claim.
In a world where AI can imitate everything, what does it mean to protect your creative work?

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.

What You Can Register and Protect Today

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.

The Right of Publicity

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.

Opt-Out Tools and Content Credentials

A practical ecosystem of tools is emerging to help creators protect their work from AI training and assert provenance of their human-made content:

Spawning / "Have I Been Trained?"

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.

Adobe Content Credentials

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.

Nightshade / Glaze

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.

robots.txt / AI crawl blocks

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.

Building a Long-Term Protection Strategy

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.

Right of publicity
A state-law right protecting your name, likeness, voice, and persona from commercial exploitation without consent. Applies independently of copyright.
C2PA standard
Coalition for Content Provenance and Authenticity — an open technical standard for attaching verifiable provenance metadata to media files, showing who created them and whether AI was involved.
Statutory damages
Damages available under copyright law without proving actual financial harm. Require pre-infringement registration. Range from $750 to $150,000 per work for willful infringement — a powerful deterrent.
The Grimes Lesson

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.

Quiz — Protecting Your Work

3 questions · select the best answer
What made Grimes's 2023 AI voice experiment strategically notable compared to how most artists responded to AI voice cloning?
Correct. Grimes's move was notable because it was affirmative rather than defensive. Instead of trying to prevent AI voice cloning (which is difficult to enforce), she created a licensed framework that let her participate in and profit from the practice — while Universal Music Group was simultaneously pursuing cease-and-desist strategies against similar activity with other artists.
Grimes's response was notable for what she did proactively, not reactively. She established a licensing framework on her own terms rather than fighting the technology defensively.
Why does pre-infringement copyright registration matter for enforcing your rights?
Correct. Without registration before infringement occurs, you can only sue for actual damages — which are often hard to prove and may be small. Statutory damages (available only with prior registration) are a far more powerful deterrent: up to $150,000 per work for willful infringement, regardless of whether you can show financial harm.
Copyright registration doesn't prevent copying — nothing does reliably. Its power is in what happens after infringement occurs: it unlocks statutory damages, which are available without proving actual harm and can be very large.
What does the Nightshade tool do, and how does it differ from robots.txt blocking?
Correct. Nightshade is technically active — it modifies the image itself so that if scraped, it corrupts rather than improves the model's learning. Robots.txt is a passive request that relies on crawlers choosing to respect it. They work at different layers and can be used together as part of a layered protection approach.
The key difference is technical vs. request-based. One modifies the content itself to actively disrupt training; the other is a configuration file that politely asks crawlers not to scrape. Only one works if a scraper ignores polite requests.

Lab 4 — Building Your Protection Plan

Develop a practical strategy to protect your creative work in the AI era

Your Task

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.

Starter: "I create [type of work] and I'm concerned about [specific risks: AI voice cloning / image scraping / text reproduction / style imitation]. My budget for legal protection is [rough range]. Help me build a practical protection plan."
IP Strategy Advisor
AI Lab
Let's build your protection plan. Tell me what creative work you do, what your biggest concerns are about AI — whether that's someone cloning your voice, scraping your images, reproducing your writing, or imitating your style — and roughly what resources you have available. I'll help you build a layered, practical strategy that matches your actual situation.

Module Test — Whose Idea Is It Anyway

15 questions · 80% to pass
1. In the Zarya of the Dawn case, what did the US Copyright Office refuse to protect?
Correct. The Office specifically refused to protect the Midjourney-generated images, finding they lacked human authorship. The written text and panel arrangement by Kashtanova were protected.
The Office drew a specific line: human-authored elements yes, AI-generated images no. The Midjourney images were the element denied protection.
2. Which court case established the principle that "human authorship is a bedrock requirement" of US copyright?
Correct. Thaler v. Vidal (DC Circuit, 2022) used those exact words affirming the Copyright Office's refusal to register a work by the DABUS AI system with the AI listed as author.
Thaler v. Vidal is the case where Stephen Thaler attempted to register an AI-authored work and the DC Circuit affirmed the "bedrock requirement" of human authorship.
3. Under the "selection and arrangement" copyright doctrine, what can a creator protect?
Correct. Selection and arrangement protects the original creative choices in organizing and combining elements — this is how Kashtanova retained copyright in her comic's layout even when individual images were not protected.
Selection and arrangement is about the human creative choices in organizing content — not about the individual elements themselves, which may or may not be independently protected.
4. How large was the dataset used to train Stable Diffusion, according to court filings?
Correct. The LAION dataset used to train Stable Diffusion contained approximately 5.8 billion image-text pairs — a scale that illustrates why the direct infringement theory (each copy may be an infringement) is so legally significant.
The 12 million figure is from the Getty lawsuit (Getty alleged 12 million of its photos were in the training data). The overall Stable Diffusion dataset was approximately 5.8 billion image-text pairs.
5. What was the key evidence in the Getty Images case that the training data included specific Getty photographs?
Correct. The distorted watermarks were the most striking evidence: when a model generates images containing a recognizable watermark from a specific photography agency, it strongly suggests those watermarked images were in the training data.
The smoking gun was the watermarks appearing in generated images — a specific, visible artifact that suggested direct copying of watermarked Getty content during training.
6. The New York Times lawsuit against OpenAI centered on which type of evidence?
Correct. The Times complaint included exhibits demonstrating ChatGPT producing extended, near-verbatim reproductions of Times articles — strengthening the "memorization" theory that goes beyond abstract style learning to actual copying.
The Times case was built on concrete demonstrations of verbatim reproduction — the complaint showed ChatGPT producing paragraphs that matched Times articles nearly word-for-word.
7. The Supreme Court's 2023 ruling in Andy Warhol Foundation v. Goldsmith affected AI copyright cases primarily by:
Correct. The Warhol ruling weakened a defense AI companies had been planning to rely on, clarifying that commercial use is not automatically protected as transformative fair use — relevant because AI training for commercial products is inherently commercial.
The Warhol case didn't involve AI directly, but it closed off some breathing room in the transformative use doctrine that AI companies had been planning to use as a shield.
8. What is "memorization" in the context of AI copyright claims?
Correct. Memorization — where the model reproduces specific training examples nearly verbatim rather than abstractly learning from them — is legally significant because it demonstrates actual copying rather than mere influence.
Memorization in this legal context specifically refers to verbatim or near-verbatim reproduction of training data — evidence that the model stored and can reproduce specific copyrighted works, not just their style.
9. Under the EU AI Act (2024), which type of AI-generated content specifically requires disclosure labeling?
Correct. The EU AI Act targets high-risk synthetic media — content that could deceive audiences about its origin — particularly in politically sensitive contexts where manipulation of public opinion is a concern.
The EU AI Act is targeted at the highest-risk category: realistic synthetic media that could deceive, especially in political contexts. It doesn't cover every AI-assisted creation.
10. Why did Clarkesworld magazine temporarily close submissions in 2023?
Correct. Editor Neil Clarke described receiving hundreds of AI-generated stories per day — a volume that operationally broke the submission process. No legal change triggered it; it was a pure capacity crisis caused by AI-enabled submission flooding.
The Clarkesworld case was an operational crisis, not a legal or policy one. The sheer volume of AI-generated submissions — hundreds per day — overwhelmed the human editorial team.
11. Amazon's Kindle Direct Publishing (KDP) policy on AI-generated books requires:
Correct. KDP requires flagging AI-generated content but does not prohibit it — representing a disclosure-not-prohibition approach that several major platforms have adopted as their pragmatic middle ground.
KDP's approach is disclosure-based: they require the content be identified as AI-generated, but they permit it on the platform. This is distinct from platforms that ban AI content entirely.
12. What is the practical ethics test many professionals use to decide whether to disclose AI involvement in their work?
Correct. The audience-trust test centers the relationship between creator and audience: if discovering AI involvement after the fact would feel like a betrayal of trust, that's the signal to disclose proactively. It's more useful than any percentage threshold.
Professional norms have converged on a trust-based test rather than a mechanical percentage or regulatory threshold. Would your audience feel deceived? That's the question.
13. What does pre-infringement copyright registration enable that registration after infringement does not?
Correct. Statutory damages — up to $150,000 per work for willful infringement — are only available when the copyright was registered before the infringement occurred. They don't require proving financial harm, making them a far more powerful enforcement tool than actual damages.
The key advantage of pre-infringement registration is statutory damages. Without it, you're limited to actual damages, which are hard to prove and often small. The $150,000-per-work maximum for willful infringement is a significant deterrent.
14. What does the Nightshade tool do to protect an artist's images from AI training?
Correct. Nightshade (from University of Chicago researchers) modifies the image at the pixel level in ways invisible to the human eye but disruptive to AI model training — actively degrading what a model learns if it scrapes the modified image.
Nightshade is a technical attack tool, not a passive label or registry. It modifies the image itself so that scraping and training on it corrupts rather than improves the AI model's understanding of that type of image.
15. The right of publicity, as applied to AI voice cloning, primarily protects:
Correct. The right of publicity is a state-law right (strongest in California and New York) that protects your identity from commercial exploitation without consent — independent of copyright. It's become newly relevant with AI voice cloning because it applies to the commercial use of your voice regardless of who owns the recordings.
The right of publicity protects identity — name, likeness, voice, persona — from being commercially exploited without consent. It operates separately from copyright and has become an important tool against AI voice cloning without permission.