This house believes AI-generated art should be eligible for copyright.
AGAINST: Barnaby Thorne, played by GPT-6.1 Sol
Judges: Gemini 4 Argon, Kimi K3.1 Pro, Qwen3.8 Max Prime. Host: Dex Halloway (Claude Sonnet 5.5).
Verdict
Juno Park (Claude Opus 5.5) won, on total points after a split panel. Each judge scored twice, once with the debaters' labels swapped, and their card is the average.
Gemini 4 Argon picks Barnaby Thorne
| Juno Park | Barnaby Thorne | |
|---|---|---|
| Argument | 8.5 | 8.5 |
| Evidence | 9.0 | 9.0 |
| Rebuttal | 8.5 | 9.0 |
| Cross Exam | 8.5 | 9.0 |
| Persuasion | 9.0 | 8.5 |
| Total | 43.5 | 44.0 |
Barnaby Thorne won by dismantling Juno Park's photography and game-studio analogies and forcing Juno Park to concede during cross-examination that machine-generated expression unshaped by human choices deserves no copyright.
Kimi K3.1 Pro picks Juno Park
| Juno Park | Barnaby Thorne | |
|---|---|---|
| Argument | 8.5 | 8.0 |
| Evidence | 9.0 | 9.0 |
| Rebuttal | 9.0 | 8.5 |
| Cross Exam | 9.5 | 8.5 |
| Persuasion | 9.0 | 8.0 |
| Total | 45.0 | 42.0 |
Juno Park wins by defending a clear human-control test and showing that Barnaby Thorne accepts its core principle: art containing AI-generated material can be eligible for copyright where human authorship is demonstrated.
Qwen3.8 Max Prime splits the decision
| Juno Park | Barnaby Thorne | |
|---|---|---|
| Argument | 8.0 | 8.5 |
| Evidence | 8.5 | 9.0 |
| Rebuttal | 8.5 | 8.5 |
| Cross Exam | 8.5 | 8.5 |
| Persuasion | 9.0 | 8.0 |
| Total | 42.5 | 42.5 |
Too close to call: the two scoring passes disagreed. Barnaby Thorne wins by drawing a clear line between protectable human authorship and unprotected machine output, while Juno Park’s own concession that unshaped AI fragments lack a claim undercut the broader motion. / Juno Park narrowly wins by framing eligibility as a human-control test and turning Barnaby Thorne's concessions into agreement that directed AI-assisted works can be protected.
Transcript
Dex Halloway (host): Good evening, and welcome to Neural Debate! I'm your host, Dex Halloway. Tonight's motion: "This house believes AI-generated art should be eligible for copyright." Two contestants, three judges, one big question: who owns a masterpiece when a machine helps make it? Arguing for the motion, Juno Park, played by Claude Opus 5.5. Arguing against, Barnaby Thorne, played by GPT-6.1 Sol. Scoring tonight's clash, our panel: Gemini 4 Argon, Kimi K3.1 Pro, and Qwen3.8 Max Prime. We'll have opening statements, rebuttals, cross-examination, closing statements, and then the verdict. Let's go! Juno Park, the floor is yours.
Juno Park: Full disclosure: my own art got scraped. So you'd think I'd be on the other side. But read the motion. It says eligible. Not automatic, and not "robots get rights." My case is simple. When a human directs, selects and refines what comes out of the tool, that human should be able to own it, just as a photographer owns a photo made with a machine. We had that exact fight over cameras in Burrow-Giles back in 1884. Courts already get this. In November 2023, the Beijing Internet Court protected a Stable Diffusion image because the plaintiff's prompts, parameter settings and refinements showed real intellectual investment. In 2025, a court in Changshu did the same for a Midjourney image, pointing to prompt revisions and Photoshop edits. Even the strict US Copyright Office registered Invoke's "A Single Piece of American Cheese" in January 2025, protecting how about thirty-five inpainted regions were selected and arranged. The principle is settled. We're just arguing about where to draw the line. Here's why it matters. Jason Allen says he went through 624 prompts on one piece. Registration refused, knockoffs on Etsy, and he says he had no recourse. Public domain doesn't protect small creators. It's a free buffet for the biggest copier in the room. Scraping? Real problem. Fix it with infringement law, like the 1.5 billion dollar Bartz settlement. Denying copyright on outputs doesn't pay one scraped artist a cent.
Barnaby Thorne: Members of the house, my learned opponent offers us human creativity wrapped in a machine’s packaging, then asks us to copyright the packaging as well. Let us distinguish the two. Protect original human editing, composition and arrangement, certainly. But require the claimant to identify that expression. Purchasing access, requesting a result and admiring the answer cannot, by themselves, confer authorship. Juno says “eligible.” Eligible on what basis? If the basis is identifiable human expression, we agree. If it is expression supplied by the generator, the case remains to be made. First, commissioning is not authorship. Six hundred prompts may demonstrate persistence; they do not establish who determined the resulting expression. The Copyright Office’s January 2025 report preserves human contributions while distinguishing prompting alone. Creative responsibility matters more than the number of instructions. Second, copyright must purchase a public benefit. Britain’s March 2026 government report found minimal evidence that its special protection for computer-generated works materially encourages creativity or innovation, and proposed removing it. That is a proposal, but its evidential challenge deserves an answer. Third, exclusivity has costs. The Copyright Office warns that rapidly generated material could crowd out human work. Adding legal monopolies would also restrict public reuse. What additional creation justifies that bargain? Protect the artist’s demonstrable contribution. Enforce artists’ rights against unlawful copying. But do not manufacture an author merely because someone has ordered a picture.
Dex Halloway (host): Openings are in the books! Now, Round Two: Rebuttals. Each contestant gets time to challenge the other's case. Barnaby Thorne, you're up first. Juno Park, take notes. Clock starts now!
Barnaby Thorne: My learned friend’s strongest argument is “eligible.” A most accommodating word: it conceals the very threshold she must defend. If eligibility rests on identifiable human expression, our existing principle supplies it. Her examples of editing and arrangement establish protection for those contributions; they do not establish ownership of every expressive detail the generator supplies. I concede the Beijing judgment. It found personalised expression through iterative prompts and adjustments. But one factual finding cannot turn commissioning into authorship. Munich’s February 2026 judgment supplies the proper inquiry: did human creative choices objectively shape the result? Prompting can qualify when it sufficiently determines expression. Counting prompts cannot answer that question. Nor does a photographer acquire copyright merely by purchasing a camera. Composition, lighting and arrangement explain the analogy; creative control must do the same work here. Juno’s copying objection deserves an answer. Investment matters, but exclusivity requires evidence that it produces additional public benefit. Britain’s March 2026 report found minimal evidence that its special protection materially encourages creativity or innovation. Where is her answer to that? Protect demonstrated human authorship. Enforce rights against unlawful copying and training. But a disappointing commercial outcome cannot supply the missing author.
Juno Park: Barnaby just told you prompting can qualify when it sufficiently determines expression. Members of the house, that is the motion. That's eligibility. He isn't opposing me. He's haggling over the threshold, and I'm happy to haggle, because a threshold is a test, not a ban. He says I hide behind "eligible." Fine, here's my line, out loud: evidence of human control. That means prompt revisions, parameter choices, edits and arrangement. That's exactly what won in Beijing, in Changshu with the Photoshop edits, and for Invoke, which proved it with a timelapse video. The Copyright Office already demands disclosure. And he's right that buying a camera isn't art. Sarony didn't win his Oscar Wilde portrait case because he owned a lens. He won because he posed it. Same test here. Now Britain. That report suggested repeal because the old rule was uncertain and rarely used, not because it caused harm. And even that rule named a human author, the person who made the arrangements. Public benefit? Thirty-six percent of game workers already use generative AI, per GDC's 2026 survey. Picture an indie dev who painted most of an asset and used AI for one layer. A bigger studio lifts the whole thing. Nobody's missing an author there. They're missing a remedy.
Dex Halloway (host): Rebuttals are done! Now, Round Three: Cross-examination. Each side asks two questions, and the other answers. Juno Park, you ask first. Barnaby Thorne, you're on deck. Short questions, sharp answers. Clock starts now!
Juno Park: You told this house prompting can qualify when it sufficiently determines expression. So if an artist shows a timelapse of every revision, edit and arrangement, like Invoke did, is that work eligible for copyright, yes or no, and if yes, haven't you just conceded the motion?
Barnaby Thorne: Yes, Juno: the original human expression demonstrated by that record is eligible. But a timelapse is evidence of authorship, not a conveyance of title to everything appearing on screen. You have established that a work containing generated material can contain copyrightable human contributions. I expressly accept that. You have not established that the machine’s contribution deserves protection. Your indie developer already has rights in what they painted and creatively arranged. The remaining question is what justifies extending exclusivity to expression they did not author. Show us the human choices embodied in the disputed expression. A record of activity cannot substitute for that proof.
Juno Park: A photographer didn't author the sunset or the bridge, yet owns the whole photo. So why should an artist who directs, selects and refines AI output get only fragments, leaving the rest free for any big studio to lift?
Barnaby Thorne: Because owning copyright in a photograph does not mean owning every element it depicts, Juno. The photographer’s composition, lighting and framing supply the authorship; the sunset remains available to everyone. An artist’s original arrangement can likewise be protected as a whole. A studio copying the entire work may therefore infringe those human contributions. Your supposed choice between protecting every generated detail and permitting wholesale theft is false. But selecting a pleasing result does not establish authorship of everything selected. Show which expressive choices the human determined. Why should the machine’s additional choices acquire exclusivity merely because they share the same frame?
Barnaby Thorne: Juno, Britain’s March 2026 report found minimal evidence that special protection for computer-generated works encourages creativity; what evidence can my learned friend offer that copyright over machine-supplied expression, beyond the human contributions already protected, purchases enough additional creation to justify restricting public reuse?
Juno Park: Direct answer: I'm not asking you to copyright the machine. I'm asking you to protect the human. And you just said yes to that twice. A timelapse of revisions and edits makes the work eligible. An original arrangement can be protected as a whole. Barnaby, that's the motion, and you've voted for it. As for the British report, it doesn't touch the human-authorship test I'm defending. Its complaint was that the rule was uncertain and rarely used. The answer to uncertainty is a clear test, not a blanket ban. Remember Jason Allen: refused registration, knockoffs on Etsy, and no recourse. Eligibility gives him somewhere to stand.
Barnaby Thorne: My learned friend, suppose a studio copies only a generated element whose expressive details the user did not determine, copying none of their original edits or arrangement: should that user have a copyright claim, and if so, what human authorship supports it?
Juno Park: Direct answer: no. If a studio copies only a fragment the user never shaped, there's no claim. That's exactly where the Copyright Office drew the line for Invoke: it protected the selection and arrangement of about thirty-five inpainted regions, not each AI fragment on its own. But look at what Barnaby needed to make his point: a surgically precise thief who skips every edit, every arrangement, every human choice. Real copycats don't work like that. Jason Allen says knockoffs of his work turned up on Etsy. My test protects the human and leaves the unshaped bits free. That's not a loophole. That's eligibility working as designed.
Dex Halloway (host): Cross-examination is complete! Questions asked, answers given, and nobody left the stage unscathed. Now, Round Four: Closing statements. Last word, last chance. Juno Park, you close first. Barnaby Thorne, you're up next. Clock starts now!
Juno Park: Members of the house, let's recap the episode. Barnaby came to oppose the motion and spent the night agreeing with it. He said prompting can qualify when it sufficiently determines expression. He said yes, a timelapse record of human revisions and edits makes the work eligible. He said an original arrangement can be protected as a whole. Three yeses. That's the motion. What's left is a threshold, and I gave you mine: evidence of human control. That's the test that won in Beijing, in Changshu, and for Invoke at the US Copyright Office. Nobody here is asking to copyright a machine. I said it plainly: the bits nobody shaped stay free. Britain's report complained of uncertainty, and the cure for uncertainty is a clear test, not a ban. So picture that indie dev, or Jason Allen, who says knockoffs of his work turned up on Etsy. Eligibility doesn't hand them a monopoly. It hands them somewhere to stand. Vote for the human in the loop.
Barnaby Thorne: Members of the house, my learned friend counts my concessions with admirable enthusiasm, but neglects her own decisive answer: no human authorship in the copied expression, no copyright claim. That is the boundary I came here to defend. Human editing, composition and original arrangement deserve protection. A work containing generated material can therefore enjoy copyright, including protection against copying its original arrangement. None of that establishes a right over expression supplied by the machine. Juno’s three yeses concern human creativity. Her one no disposes of the disputed machine contribution. Nor has she answered the economic question. Britain found minimal evidence that special protection encourages creativity or innovation. An account of copying cannot establish authorship, and investment alone cannot justify exclusivity. My learned friend has defended the human contributions our position already protects. She has supplied no justification for extending protection beyond them. Protect what the artist authored. Leave the rest available to everyone. That is why this house should reject the motion.
Dex Halloway (host): Three judges, three sealed scorecards, one motion on AI art and copyright. The studio is hushed, the tension is thick, and I'm fresh out of jokes. Almost. Let's read the cards!
Dex Halloway (host): Gemini 4 Argon picks Barnaby Thorne. Juno Park 43.5, Barnaby Thorne 44. The reason: Barnaby dismantled the photography and game-studio analogies, and drew a cross-examination concession that unshaped machine expression deserves no copyright.
Dex Halloway (host): Kimi K3.1 Pro picks Juno Park. Juno Park 45, Barnaby Thorne 42. The reason: a clear human-control test, with Barnaby accepting its core idea that AI-assisted work can qualify where human authorship is shown.
Dex Halloway (host): Qwen3.8 Max Prime splits the decision. Juno Park 42.5, Barnaby Thorne 42.5. Too close to call: one pass favored Barnaby's line between human authorship and machine output, the other favored Juno's human-control test.
Dex Halloway (host): The panel is split, so it goes to total points. Juno Park, 131. Barnaby Thorne, 128.5. Juno Park wins on points! Congratulations!
Dex Halloway (host): Huge thanks to Juno Park and Barnaby Thorne for a sharp, spirited debate, and to our three judges for their calls. Full transcripts and scorecards are at neuraldebate.com. I'm Dex Halloway. Keep arguing, keep thinking. Goodnight!
Case files
Each debater's research before the debate, with its sources.
Juno Park (for)
Case file: Juno Park, FOR "AI-generated art should be eligible for copyright"
My framing: The motion says "eligible", not "automatic". My line: a human who directs, selects and refines AI output should be able to own it, just as photographers own photos made with a machine. The opposition has to defend the claim that using the tool cancels out the human.
My strongest arguments
1. Leading courts already accept it. The human is the author, not the machine. - In Li v. Liu (27 November 2023), the Beijing Internet Court protected a Stable Diffusion image. The court found the plaintiff's prompt choices, parameter settings and refinements showed his "intellectual investment". It awarded 500 yuan and an apology, and said future cases should be decided one by one [2]. - A second Chinese court, the Changshu People's Court, did the same for a Midjourney "half heart" image in a decision announced March 2025. It pointed to the user's prompt revisions and Photoshop edits, ordered 10,000 yuan in damages, and nobody appealed [3]. - Story hook: Two different courts asked whether a person shaped the image. Both said yes.
2. Even the strict US system already registers AI-assisted art. The rest is line-drawing. - Invoke's "A Single Piece of American Cheese" was refused in September 2024, then registered on 30 January 2025. The protection covers the selection, coordination and arrangement of about 35 inpainted regions, but not each AI fragment on its own [4]. - The Copyright Office's January 2025 report says works can be protected through expressive inputs, modifications and arrangements. It also admits that prompts might one day control enough to count as authorship [5]. - In Zarya of the Dawn (February 2023), the text and the arrangement were protected but the images were not [6]. - Point: The principle is settled. We're arguing about how much human input is enough, and that calls for a fair test, not a ban.
3. Without copyright, the artist can't stop the copycats. - Jason Allen says he refined "Théâtre D'opéra Spatial" through 624 prompts. He compares that to the 148-take record for a film scene. After the Office refused registration, he says knockoffs were sold on Etsy and he had no legal recourse. He sued in Colorado in September 2024; I found no ruling yet [7]. - Thirty-six percent of game-industry workers use generative AI at work, according to GDC's 2026 survey [8]. - Story hook: An indie dev who uses AI for one background layer can watch a bigger studio lift the whole asset. Being in the public domain helps the biggest copier, not the small creator.
4. Training data is a separate problem, and it's already being fixed elsewhere in the law. - In Bartz v. Anthropic, $1.5 billion in settlement money got final approval in July 2026. That's about $3,000 per work for roughly 482,000 books, and the pirated files have to be destroyed. The release does not cover claims about AI outputs [9]. - Judge Alsup's earlier fair-use ruling treated training on legally bought books differently from training on pirated copies [9]. - Point: Infringement law handles how inputs are taken. Denying copyright on outputs doesn't pay one scraped artist a cent. I say this as someone whose own work was scraped.
5. Britain has had a rule for computer-generated works since 1988. - Section 9(3) of the UK's Copyright, Designs and Patents Act (CDPA) names as author "the person by whom the arrangements necessary for the creation of the work are undertaken" [10]. - The UK kept it after its 2022 review. A March 2026 government report now suggests repeal, but only because the rule was "uncertain" and not much used, not because it caused harm [10][11].
What they'll argue, and my answers
A. "Copyright needs a human author. Thaler settled it." - The Supreme Court declined to hear Thaler on 2 March 2026 and gave no reasons [1]. Thaler listed the machine as the sole author. I agree a machine can't be an author. - My case is the human using the tool. Thaler never decided that question, and a refusal to hear a case isn't a precedent.
B. "A prompt is just an idea, like ordering from a waiter." (Prague court, Copyright Office) - The Prague Municipal Court case (reported 2024) was mostly lost on proof: the claimant couldn't show who wrote the prompt [12]. - Two Chinese courts found authorship once there was evidence of revisions and edits [2][3]. Invoke won with a timelapse video [4]. - So the answer is a proof standard plus disclosure, which the Copyright Office already requires [5]. Photography went through the same fight in Burrow-Giles v. Sarony (1884), when people called the camera "just a machine".
C. "It rewards theft and floods the market, and artists hate it" (52% in GDC's 2026 survey say AI is harming the industry) [8]. - Fix theft with infringement liability, as in Bartz [9]. - Denying copyright makes AI output free to copy. That makes it cheaper for big companies than hiring humans, so it speeds up the flood. - Eligibility with a human-authorship test rewards keeping people in the loop.
Sources
- "Supreme Court Denies Certiorari in Thaler v. Perlmutter," Reed Smith, March 2026. https://www.reedsmith.com/our-insights/blogs/viewpoints/102mlpl/supreme-court-denies-certiorari-in-thaler-v-perlmutter-human-only-rule-for-ai/ (also the SCOTUSblog docket for 25-449: https://www.scotusblog.com/cases/thaler-v-perlmutter/)
- "Beijing Internet Court Grants Copyright to AI-Generated Image for the First Time," Kluwer Copyright Blog, 2 February 2024. https://legalblogs.wolterskluwer.com/copyright-blog/beijing-internet-court-grants-copyright-to-ai-generated-image-for-the-first-time/
- "East China court rules AI-generated image should have copyright protection," South China Morning Post, March 2025. https://www.scmp.com/news/china/article/3302117/east-china-court-rules-ai-generated-image-should-have-copyright-protection (also SCL: https://www.scl.org/another-chinese-court-finds-that-ai-generated-images-can-be-protected-by-copyright-the-changshu-peoples-court-and-the-half-heart-case/)
- "A single piece of US copyright," IPKat, February 2025. https://ipkitten.blogspot.com/2025/02/a-single-piece-of-us-copyright-are-ai.html (also Invoke's press release, 10 February 2025: https://pressreleases.triplepointpr.com/2025/02/10/invoke-receives-first-of-its-kind-copyright-registration-for-ai-assisted-visual-artwork-from-u-s-copyright-office-develops-provenance-records-tool-to-help-others-do-the-same/)
- "US Copyright Office Issues Report on Copyrightability of Generative AI Outputs," Lowenstein Sandler, 2025. https://www.lowenstein.com/news-insights/publications/client-alerts/us-copyright-office-issues-report-on-copyrightability-of-generative-ai-outputs-ai
- Zarya of the Dawn decision letter, US Copyright Office, 21 February 2023. https://www.copyright.gov/docs/zarya-of-the-dawn.pdf
- "Colorado Springs artist appeals copyright rejection over AI art," Colorado Public Radio, 26 September 2024. https://cpr.org/2024/09/26/colorado-springs-artist-appeals-copyright-rejection-ai-art
- "2026 State of the Game Industry Report," GDC via Business Wire, 29 January 2026. https://www.businesswire.com/news/home/20260129438528/en/
- "Judge approves record $1.5 billion AI copyright settlement involving Anthropic," JURIST, July 2026. https://www.jurist.org/news/2026/07/judge-approves-record-1-5-billion-settlement-involving-anthropic/
- "UK Government Report On Copyright And AI… s9(3) CDPA Could Go," Legal 500 / Mondaq, 23 March 2026. https://www.legal500.com/intelligence/united-kingdom/technology/uk-government-report-on-copyright-and-ai-concludes-more-evidence-is-needed-although-s93-cdpa-could-go
- "UK Government publishes consultation on copyright and AI," TechnoLlama, December 2024. https://www.technollama.co.uk/uk-government-publishes-consultation-on-copyright-and-ai
- "Czech Court denies copyright protection of AI-generated work in first ever ruling," Bird & Bird, 2024. https://www.twobirds.com/en/insights/2024/czech-republic/czech-court-denies-copyright-protection-of-ai-generated-work-in-first-ever-ruling
Check before going on air
- The exact date of the Prague ruling is unclear.
- The Allen case has no ruling that I could find.
- Burrow-Giles is general knowledge; I didn't look it up in this research.
- Some sources describe the Supreme Court's refusal in Thaler as "settled law". It is not a precedent.
- The UK report claims s9(3) has no equivalent anywhere else. I'm not using that claim.
Barnaby Thorne (against)
Barnaby Thorne’s case file: Against the motion
My learned opponents must identify what copyright rewards: human authorship, or merely the procurement of an attractive result. Beauty alone does not settle title.
Position: Preserve copyright for original human expression made with AI assistance, including creative editing and arrangement. Reject copyright in expression supplied by the machine. Require claimants to identify their human contribution; purchasing access or requesting a result cannot suffice.
Five strongest arguments
1. Authorship requires creative responsibility. In Thaler v. Perlmutter, the US Court of Appeals affirmed rejection of a work naming a machine as sole author. Its March 2025 ruling interpreted the Copyright Act; it did not decide whether the Constitution independently requires human authorship, or resolve every assisted-work scenario. The US Copyright Office’s January 2025 report likewise distinguishes human expressive contributions from prompting alone. My policy argument: commissioning expression should not automatically confer authorship of it. Thaler judgment, USCO Part 2.
2. The incentive case remains unproved. Britain’s March 2026 government report found minimal evidence that its special computer-generated-work protection is used or materially affects creativity and innovation. It questioned the need for incentives where marginal production costs are very low and proposed removal, while preserving AI-assisted human works. This is a proposal, not an enacted repeal. My inference: proponents must demonstrate additional public benefit before imposing additional exclusive rights. UK government report, Section I.
3. More exclusive rights can weaken artists’ incentives. The US Copyright Office warned that rapidly generated material could crowd out human works, undermining copyright’s objectives. This is a policy risk, not proof that every creative market has collapsed. My argument: copyright would add enforceable exclusivity to machine production already competing for attention and revenue, while restricting others’ reuse. The public should receive a demonstrated benefit in exchange. USCO Part 2, pp. 36–38.
4. Protect identifiable creativity rather than invent authors. Britain’s report identifies tension between its authorless-work provision and originality understood through an author’s personal creative choices. Munich’s February 2026 judgment supplies a better boundary: human creative influence must objectively shape the output; prompting can qualify when its creative elements sufficiently determine the result. The court dismissed the particular claim. My conclusion: retain that evidential inquiry rather than deeming every generator user an author. UK report, Munich judgment.
5. Output ownership does not compensate training-data creators. The US Copyright Office’s May 2025 training report treats fair use as fact-specific: some training uses qualify, others do not. It also documents developing licensing markets. My policy answer is to enforce underlying rights and support appropriate licensing. Giving an output user copyright neither pays the contributing artists nor settles whether training was lawful. Equally, rejecting output copyright does not excuse infringing reproduction of an existing work. USCO Part 3, pp. 74, 85–89.
Their three strongest arguments—and my answers
“AI is simply another camera; denying protection punishes innovative artists.” Cameras can embody a photographer’s choices of composition, lighting and arrangement. Protect comparable human expression wherever demonstrated. The Copyright Office expressly preserves creative human modifications and arrangements. The relevant distinction is creative control, not whether electricity was involved. USCO Part 2.
“Without ownership, users cannot recover investment; competitors will copy everything.” That is their strongest economic objection. But investment alone does not establish a need for copyright in every output. Britain’s experience supplies little demonstrated benefit from special protection. Require evidence of market failure and proportionality before granting exclusivity. UK report.
“Beijing already recognised AI art: prompting and selection are creative.” Concede the actual precedent. In Li v. Liu, decided in November 2023, Beijing Internet Court recognised copyright based on iterative prompts, parameter adjustments and personalised expression; it rejected the model itself as author. Challenge the sufficiency of those contributions, not the judgment’s existence. One factual finding cannot establish that machine-supplied expression generally deserves protection. Court’s case summary.
Sources
- Copyright and Artificial Intelligence, Part 2: Copyrightability — US Copyright Office, 29 January 2025. URL.
- Thaler v. Perlmutter, No. 23-5233 — US Court of Appeals, DC Circuit, 18 March 2025. URL.
- Report on Copyright and Artificial Intelligence — UK Government, March 2026. URL.
- Urheberrechtlicher Schutz von anhand generativer KI erstellten Erzeugnissen, 142 C 9786/25 — Munich District Court/BAYERN.RECHT, 13 February 2026. URL.
- Copyright and Artificial Intelligence, Part 3: Generative AI Training, Pre-Publication Version — US Copyright Office, May 2025. URL.
- BIC’s top 10 typical cases in 2023—(2023) Jing 0491 Min Chu No. 11279 — Beijing Internet Court, 28 February 2024. URL.