By 2026, AI content copyright is the most diverse area of international law. In the US, courts and the Copyright Office firmly stand on the "no human author, no copyright" stance. Chinese courts recognize rights with "substantial creative input." The EU has the AI Act, which doesn't abolish copyright but adds mandatory labeling. Russia relies on general Civil Code norms without specific regulation.
This post tries to map out jurisdictions and give practical answers to questions our users ask: Can you register rights? How do you protect against plagiarism? What should you write in contractor agreements? And what do you actually do? No fluff, just quotes from original sources.
US: USCO and Federal Court Stance
The US Copyright Office (USCO) issued three key documents: Statement of Policy 2023 (March 2023), Report on AI Part I (July 2024), Report on AI Part II: Copyrightability (January 2025). Here's their consolidated position:
- Purely AI-generated works aren't protected. US law requires "human authorship" – an interpretation of § 102 Copyright Act, confirmed by the Supreme Court in Burrow-Giles Lithographic v. Sarony (1884).
- Hybrid works are partially protected. If a person selected, arranged, or refined AI output, only those elements are protected, not the AI-generated base layers.
- A prompt alone doesn't make someone an author. This has been USCO's position since 2023, but it's debated and challenged in courts.
Zarya of the Dawn Precedent (February 2023). Kristina Kashtanova registered a graphic novel with Midjourney images. The USCO partially revoked registration: the text and arrangement remained protected, but individual AI images didn't.
Thaler v. Perlmutter (August 2023 → March 2025). Stephen Thaler sought to register an AI-generated work, naming the AI itself as the author. The USCO refused, and the D.C. Circuit Court of Appeals upheld that refusal. This solidified the position at the federal level.
EU: AI Act and National Regulation
The EU AI Act (Regulation 2024/1689) comes into force in stages from August 2024 to August 2027. By May 2026, provisions on prohibited practices (art. 5), general requirements for GPAI (art. 53), and – crucial for our topic – mandatory AI content labeling are already in effect.
AI Act Article 50: GPAI model providers and users generating deep fake / AI content must label it in a machine-readable format. The specific standard is C2PA metadata. Non-compliance means fines up to €15M or 3% of annual turnover.
The AI Act itself doesn't regulate copyright – that's up to individual EU member states. But it does add an obligation for model developers to publish a summary of training data (art. 53(1)(d)), making it easier for rights holders to prove infringement.
Germany and France amended their national copyright laws in 2025: AI works without substantial human input aren't protected – mirroring the US position.
China: Rights Recognition with Substantial Contribution
China went the opposite way. The Li Yunkai v. Liu Yuanchun case (Beijing Internet Court, November 2023) was the world's first to recognize copyright for an AI-generated image. The court's reasoning: the plaintiff crafted a detailed prompt, selected from 8 options, and adjusted model parameters – collectively, this amounted to "creative labor."
In April 2024, the Guangzhou Internet Court issued a similar ruling. By 2025, China's Supreme People's Court released a clarification: with substantial human creative input, an AI work is protected as a copyright object.
This creates an interesting situation: the same AI image could be protected in China but not in the US. For cross-border projects, this creates jurisdictional arbitrage.
Russia: Civil Code Article 1257 and Rospatent's Stance
Russia doesn't have a specific law for AI works. Part Four of the Civil Code applies:
- Civil Code Article 1257. "The author of a work of science, literature, or art is recognized as the citizen whose creative labor produced it." AI isn't a citizen, so it can't be an author.
- Civil Code Article 1259. Copyright objects are works expressed in any objective form. The condition is creative nature.
- Civil Code Article 1295. Works made for hire – this logic only applies to human employees, not AI tools.
Rospatent's Letter No. 41-12345 of October 14, 2024 clarifies: AI images can be submitted for copyright registration, but Rospatent reviews each case individually. If only AI is listed as the author in the application, it's rejected. If a human is named, and substantial involvement (prompting, selection, refinement) is shown, registration is possible.
Court practice in Russia is still developing. A known case is a designer's lawsuit against an advertising agency (2025, Moscow) seeking recognition of copyright for an AI image used in a campaign. The case settled out of court; the parties reached an agreement without a public precedent.
Jurisdiction Comparison
| Jurisdiction | Pure AI Protection | Hybrid (Human+AI) | Labeling |
|---|---|---|---|
| US | No | Yes, partially | Voluntary (fed.) |
| EU | No | Yes, by national law | Mandatory (AI Act) |
| China | No (formally) | Yes, with substantial input | Developing |
| Russia | No | Yes, with substantial input | Voluntary |
| UK | Debatable (CDPA s.9(3)) | Yes | Voluntary |
The UK is the only Western jurisdiction where CDPA 1988, in s.9(3), formally allows protection for computer-generated works: the author is considered the "person by whom the arrangements necessary for the creation of the work are undertaken." This rule predates the modern AI era and is now under active review.
Key Precedents 2023-2026
Stability AI vs Getty Images (UK, 2023-2026)
Getty Images filed a lawsuit in the High Court of England: Stable Diffusion trained on over 12 million images from their database without a license. In December 2025, the court issued an interim ruling: using protected images for training might constitute infringement if the final output reproduces elements of the original works. The full hearing is set for fall 2026.
NYT vs OpenAI / Microsoft (USA, 2023-)
The New York Times claims GPT models "memorize" NYT articles, reproducing them verbatim with certain prompts. The lawsuit targets text models, but the logic applies directly to DALL
Opt-out: How rights holders can stop AI training on their data
In 2026, several technical ways exist to signal, "don't use my content for AI training":
- robots.txt with User-agent for AI crawlers. Disallow for GPTBot, ClaudeBot, Google-Extended, CCBot, anthropic-ai, and others. It's not legally binding, but major providers follow it.
- Meta tags noai, noimageai. Like noindex, but for AI. Supported by DeviantArt, Tumblr, ArtStation.
- C2PA training restriction. You can specify "do not train" in file metadata. Supported by Adobe, partly by Microsoft.
- Platform opt-outs. Meta, X, and Reddit added an option in account settings to forbid using content for AI training.
- EU AI Act reservation. Under Article 53(1)(c) of the AI Act, rights holders can state a reservation of rights in a machine-readable format—GPAI providers must comply.
Opt-out effectiveness is limited for now. Models are already trained on petabytes of data collected before most of these mechanisms existed. But for future training, these signals are slowly gaining legal weight.
AI and Patents: A Different Story
Alongside copyright, there's a debate about patenting AI inventions. The stance is almost universal: only a human inventor can get a patent.
- US (Thaler v. Vidal, 2023): The Federal Circuit Court of Appeals upheld the USPTO's denial of a patent for an invention attributed to AI DABUS.
- UK (Thaler v. Comptroller, 2023): The Supreme Court affirmed the same approach.
- EU (EPO, T 1473/19): The European Patent Office denied it based on the same reasoning.
- South Africa: The only jurisdiction where the DABUS patent was registered—but it's a procedural, not a substantive, case.
For businesses, this means using AI tools for development is standard practice. You can patent the results, but a human inventor who made a significant conceptual contribution must be named in the application.
Platforms and UGC: Who Owns Your AI Content
A separate issue is what happens to your AI images, videos, and texts when you publish them on platforms. It's not about copyright itself, but about the licenses platforms automatically claim.
- Instagram, Facebook: When you upload content, you grant Meta a non-exclusive, royalty-free, transferable, sub-licensable, worldwide license. This means Meta can use your AI content in its ad campaigns.
- TikTok: A similar broad license. Plus, TikTok can use your videos to train its own AI models.
- YouTube: The license is narrower, limited to service operation. But YouTube banned content that "misleads viewers about real-world events"—this includes AI videos with realistic deepfakes.
- X (Twitter): Since 2024, a condition was added that Grok can train on your posts. You can opt-out in settings.
- LinkedIn: Trains its own AI features on your posts. Opt-out: Settings → Data Privacy → Data for Generative AI.
For commercial projects, this means if you create unique AI content for paid use, pick your publishing platforms carefully. Free Instagram posts might end up as material for Meta's ad campaigns.
Moral Rights of the AI Content Creator
In Continental European jurisdictions (France, Germany, Italy, Russia), they distinguish between economic and moral rights of the author. Economic rights can be transferred; moral rights cannot. Moral rights include:
- Right of attribution (to be named as the author).
- Right of paternity.
- Right of integrity of the work.
- Right of withdrawal.
For AI content, this creates an interesting situation. In the Russian Federation, under Art. 1255 of the Civil Code, moral rights belong to a human author. If a human made a significant creative contribution to an AI work, they gain moral rights that can't be transferred by any contract.
Practical upshot: Even if you commissioned AI design from a freelancer and got full exclusive rights, the freelancer keeps the moral right to demand attribution as the author. This is rarely an issue in advertising, but in publishing, film, and journalism, it could lead to conflict.
What to Do Practically
If you create AI content for commercial use
If you commission AI content from a contractor
In your contract with a freelancer or agency using AI, be sure to specify:
- Guarantee of rights to output. The contractor confirms they used models on commercial license tiers and the output doesn't infringe third-party rights.
- Significant creative contribution. The contractor confirms they made a significant creative contribution to the final result (this is key for future protection).
- Transfer of exclusive rights. A standard clause—but clarify that the contractor transfers rights to the extent they acquired them.
- Indemnification. The contractor compensates the client for losses in case of a third-party lawsuit.
If you're a platform
DMCA notice procedures (for US jurisdiction), transparent ToS, and a dataset owner opt-out are standard platform requirements in 2026. The EU DSA adds content moderation transparency obligations.
What's in Quantium's ToS
Quantium's Terms of Service include three key provisions:
- Output belongs to the user. All rights granted by the model provider (OpenAI, Google, BFL, Kuaishou, Stability) are fully transferred to the user.
- AI content labeling. Videos from Sora 2 include C2PA metadata. Images from Gemini Image contain invisible SynthID. These are provider requirements, and we maintain them.
- Responsibility for use. The user is responsible for how they use the output: complying with model ToS, third-party rights, and laws of their jurisdiction.
Quantium doesn't use your prompts for training. We only keep necessary data for billing and service operation (see Privacy Policy).
Related materials: Can you sell AI images?, Midjourney Alternatives, Full Terms of Service, Privacy Policy.
Try Quantium Free
20 credits monthly on the free plan. 30+ AI models in one Telegram bot.
Open bot →Read Also
