Glossary
What is copyright and AI
In short
Copyright and AI is the set of questions about who owns texts and images made by a neural network, and whether models may train on other people’s works.
Security and law · copyright
MMOOZZGG.com editorsChecked: 2 min read
For a child
If you draw a postcard, it is yours: you chose the colors and who is on it. When a smart program draws a postcard from one request of yours, grown-ups argue about whose it is. They also argue whether such programs may learn from other people’s drawings and books without asking. So it is important to respect other people’s work and to ask for permission.
In plain words
The law protects a work that a person created by creative effort. A neural network is not a person, so an image made from a short prompt may get no protection at all. If a person reworked the result a lot, or chose and arranged the parts, that contribution is protected. The second question is training: models learn from huge sets of books, articles and images, and authors argue whether they must be asked and paid. The third question is the result: if a neural network gives an almost exact copy of someone else’s work, publishing it is risky, because it may break the rights of the original author.
For business
A company faces three risks here. First, a logo, text or image made by a neural network with almost no human input may get no protection, and it will be hard to stop others from copying it. Second, the result may repeat someone else’s work or trademark. Third, the terms of the service decide who owns the result and whether you may use it commercially. In Russia, from 1 March 2027, Federal Law 243-FZ requires services built on large models to tell users who owns the rights to the result and on what terms it is available. Until then, read the terms of your service, leave the creative finishing of key materials to a person and check results for similarity to known works. Put the rules for working with generated content into the staff guidelines.
For a specialist
Under Articles 1228 and 1257 of the Civil Code of the Russian Federation, the author is the citizen whose creative work produced the result; a person who gave only technical help does not become an author. So a model cannot be an author, and protection of a generated result depends on the human creative contribution. The U.S. Copyright Office took a similar position in its report on generative AI (January 2025): protection does not cover purely generated material, prompts alone are usually not enough with today’s technology, and human selection, arrangement and modification are protected. For training, from 1 March 2027 Russian Federal Law 243-FZ (Article 10, Part 2) does not treat as a violation the analysis and short-term reproduction of works to train sovereign and national models, if the copy was obtained lawfully or the work is available without technical restrictions. In the EU, Article 53 of Regulation 2024/1689 requires providers of general-purpose models to respect rights holders’ opt-outs from text and data mining and to publish a summary of the training data.
Example
A studio created a character for a brand mascot with a neural network. The lawyer asked the designer to redraw it by hand and keep the history of edits: this way the company gets a work with a human contribution. The drawing itself was checked for similarity to well-known characters.
Related terms
Questions and answers
Can you train a neural network on other people’s texts?
Sources
- Civil Code of the Russian Federation, Part Four, Article 1228, ConsultantPlusconsultant.ru
- Civil Code of the Russian Federation, Part Four, Article 1257, ConsultantPlusconsultant.ru
- U.S. Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability, January 2025copyright.gov
- Federal Law No. 243-FZ of 26 July 2026, Article 10, ConsultantPlusconsultant.ru
- Regulation (EU) 2024/1689, Article 53, AI Act Service Desk of the European Commissionai-act-service-desk.ec.europa.eu
Not sure whether your company needs this?
AI solutions map