Posted inAuthor's

Copyright and Generative AI in 2025: A Complete Guide for Russian Creators and Companies

Artificial intelligence has gone from being a handy tool to a manufacturing infrastructure: it writes, draws, voices, compiles code, improves image and sound, cleans data and helps with production processes. In almost every scenario, copyright emerges: who is the author of the result, whether it is possible to train the model on other people’s works, when it is permissible to «borrow» and how to issue rights if AI was involved in the project. Below is an expanded, practice-oriented guide focusing on Russian jurisdiction and taking into account current international trends and cases.

1) Map of law: how security is arranged in Russia, the USA and the EU

Russia (CG of the Russian Federation, part four).
The «works of science, literature and art» in any form, regardless of merit and mode of expression, are protected. creativeness and human authorship: the author is a citizen whose creative work created the work, it is laid down in the norms on the subject and object of copyright (Articles 1228, 1257, 1259 of the Civil Code of the Russian Federation), from this follows the basic position: AI by itself cannot be an author.It is also important to remember that purely machine-generated «inferences» without human creative input are not usually protected as works, and it is also important to remember that ideas/methods/principles are not protected (Article 1259, paragraph 5 of the Civil Code). ConsultantPlus+1

Registration in Russia for copyright works required: right arises at the moment of creation. computer programs and databasesThey can (and often usefully) be registered with Rospatent – this makes it easier to prove the rights and output versions. Russian Patent Office +1

United States (human authorship, fair use).
In the US, the regulator’s position is detailed in the guidance and reports of the Copyright Office (USCO): “pure” AI-generated materials unregisteredprotected discernible human contribution In January 2025, the second part of the USCO report on “copyrightability” was released, which systematically reinforces this approach, and in May 2025, a preliminary version of Part 3 of the report on “copyrightability” was published. teaching AI on protected works (fair use assessed) case-case It is not an “automatic indulgence”. copyright.gov+2copyright.gov+2

Doctrine fair use 17 U.S.C. §107 – Four factors: purpose of use (including commercial nature), nature of the original, volume and materiality of borrowing, market impact, and these factors apply to each dispute in particular, and are the basis of ongoing lawsuits against AI developers. Legal Information Institute +1

EU (TDM exclusions and transparency of education)
The Digital Single Market Directive (Directive (EU) 2019/790, “DSM”) introduced exemptions for the use of digital single market (DSM) Text and Data Mining (TDM):
— Art. 3 — for the R&D of cultural heritage organizations and institutions;
— Art. 4 — broad TDM, provided that the copyright holder I didn’t opt-out. This is done in a machine-readable manner, and is implemented in the national laws of the EU countries. AI Act (EU Regulation 2024/1689) fixed transparency General AI Models (GPAI): Publication summary description of training dataand in July 2025 published a mandatory pattern This is not the same as «dataset distribution,» but it increases the verifiability of the sources. bundesnetzagentur.de+3Eur-Lex+3Eur-Lex+3

Conclusion for the Russian reader.
There is no universal “fair use” in Russia; exceptions are closed and interpreted narrowly.
To teach models on other people’s works Best way — rights/licenses or legal hygiene (sources with which it is permissible).
When using foreign services or working with the EU/US, you are subject to their regimes (TDM-opt-out, transparency templates, injunctions/retentions, etc.). ConsultantPlus

(2) When AI content can be protected: human input, evidence and case studies

1) “Net generation” (prompt → result).
If your contribution is reduced to a short hint without creative selection/componenting, there is no right in Russian logic (no human author), in the US such material is not registered (USCO), indicating that there is no “human authorship”. copyright.gov

2) AI as an author’s tool (assistance).
Guarded. your creative input: selection, editing, collage, drama, storyboarding, layout of musical themes, working with timing, etc. In the United States, when registering, you need to disclose the share of AI; in Russia, registration is not necessary, but document the processTo confirm what you have done:
— Save drafts/versions, screencasts and key prompts;
Changelog (what and why you changed it)
— Store the source (clip packages, tracks, scenes).

(3) Code and databases.
Software and databases can be registered in Rospatent — this strengthens the evidence position (public registry, date / volume, copyright holder). creator (and/or work-related employers) not AI. Russian Patent Office

(4) Which is not protected.
Ideas, methods, algorithms as such, “styles in general” – outside copyright protection (Paragraph 5 of Art. 1259 of the Civil Code of the Russian Federation). ConsultantPlus

Practical checklist of the author / editorial board:
— Get it done. AI in the editorial office / studio (which is permissible, not, how we mark);
Record the human contribution and keep the evidence;
In contracts with authors/contractors, require disclosure of AI participation and transfer of logs/sources;
When working “based on” living artists/musicians, assess the risks of “recognisable imitation” (reputation, contractual prohibitions, in some countries – individual claims).

(3) Training models on other people’s data: permissions, exclusions, practices and risks

Licenses are the “gold standard.”
When rights holders themselves This is a legitimate way to allow content to be used in learning (media houses, runoffs, aggregators) and in the last two years, major publishers have been making such agreements (e.g., Axel Springer; others through partnerships/arrangements) that reduce entry disputes and increase product sustainability. axelspringer.com+1

Courts and public claims.
A classic example is a lawsuit. The New York Times The court allowed the process to continue, in 2025, orders were issued for data retention (output logs) and other procedural issues; the essence of the dispute is the use of protected materials in training and “literal” outputs. Ziff Davis (CNET/PCMag/IGN/Lifehacker portfolio) has also filed a training and inference lawsuit. The results have not yet reached a substantive conclusion — disputes are moving, and court decisions are being carried out. factorial character (by factuality of specific sets and conclusions). Reuters+3Reuters+3Reuters+3

UK: Autumn Decision 2025 in Case Getty Images v. Stability AI The High Court of England and Wales has found itself “mixed”: most copyright claims have effectively “failed” due to jurisdictional issues and lack of evidence of training in the UK. trademark The court found violations, and the court noted separately that model as such It does not store «copies» of works in the sense of copyright, which has become an important but narrow conclusion precisely in the context of this process. parallel to the Getty continues the dispute in the United States. Conclusion: resonance is high, but «general formula» of the legality of learning this case. yet. Reuters+2 Guardian+2

Fair use RUB «automatic indulgence».
Even in the US, where there is fair use, USCO’s Part 3 of the report emphasizes the value of each factor and the particular importance of market impact (loss of sales/licenses, crowding out markets). copyright.gov+1

EU: TDM and transparency.
If you are working with EU users/data, check whether the right holder has included the rightholder. opt-out (Article 4 DSM), whether you comply with storage and disclosure regimes, and for GPAI, whether published training report AI Office template (from July 2025). Eur-Lex+2OUP Academic+2

Russia: What does this mean in practice?
“Free” learning on protected works in Russia no (without any separate basis).
Risks: claims of infringement (reproduction/processing/bringinging to the public), blocking according to the requirements of the right holders, reputational and contractual consequences (including for customers).
Dataset hygiene: sources with clear licenses, exclusion mechanisms at the request of the copyright holder, filters for obvious violations (including the removal of watermarks / logos / recognizable “dictionaries” of specific authors from the output), audit of traces of “remembering”. ConsultantPlus

Mini-check compliance list for developers / product:
— register of sources and licenses; versions of datasets; date of «cutting»;
— automatic opt-out-Filters and respect robots.txt;
— the protocol of removal at the request of the right holder;
policy no-train for customer data (by default) and separate “clean” environments;
— a journal of training runs and control of «leaks» (tools against «literal» reproductions).

(4) Implementation practices: AI policies, contractual clauses, evidence and workflows

A. Your internal AI policy (for media, studios, product teams):

  1. TransparencyLabel materials where AI has had a significant role; for B2B, keep a hidden origin label.
  2. Allowed scenariosList “green” (noise reduction, upscale, localization), “yellow” (generation “based on motives”, stylization of living authors) and “red” (imitation of a specific author / brand without permission).
  3. Logs and versions: Obligate all participants to keep sourceIntermediate iterations and prompts.
  4. Checking rights: before release — checklist for images / fonts / music / trademarks / personas.
  5. Customer dataBy default, no-trainSeparate consent in writing.

B) Contractual reservations (slices-required).
Disclosure of AI Participation

The Contractor confirms that when creating the Work, he used (a) / did not use (a) artificial intelligence tools. In case of use, the Contractor provides the Customer with a list of the used funds, a description of the Contractor’s creative contribution and source materials (prompts, intermediate versions, logs of changes).

Guarantees and purity of rights

The Contractor guarantees that the materials provided to the Work (a) and used in its creation do not violate the rights of third parties, including copyright and related rights, trademark rights and images of persons. The Contractor ensures the release of the Customer from claims of third parties (indemnity) within the agreed limit of liability.

No-train (personal/client data)

The Parties have agreed that Customer data, Work materials and associated metadata are not used for additional training, fine-tuning or improving AI models without the Customer’s separate written consent.

Licensing of datasets

For training datasets, Parties shall confirm that sufficient rights/licenses are available or that they comply with applicable exceptions, and a register of sources, terms of use and date of update shall be provided upon request.

Marking and transparency for the EU (if necessary)

In the event of an EU market launch, the Contractor shall publish a summary description of the training data on the current AI Office template within the time and scope set out in Regulation (EU) 2024/1689 and shall ensure respect for the opt-out of rightholders pursuant to Article 4 of Directive 2019/790 (DSM).

(C) Evidence: how to “lay a straw”.
— depositing key releases;
hashes and timestamps of source / version;
— storage of the history of prompts / iterations;
clear rights within the team (who contributed what)

D) Model “red flags”.
“literal” or almost literal pieces of known texts/tracks/photos;
— the appearance of other people’s logos / watermarks in the output of the model;
Styling “like X” in advertising tasks without a license;
— absence of sources and traces of human work.

E) Russian specificity (it is important to remember).
The author is human; the AI is a tool, and it affects what you protect and how you protect it. ConsultantPlus
Registration of software / database in Rospatent is a quick way to strengthen the position (it will be useful in disputes and transactions). Russian Patent Office
Fair use is not valid; any reference to American practices is nothing more than a reference point, not an indulgence. Legal Information Institute

Where to look so as not to “lose the thread” (norms and key documents)

  • CC RF, Part IV — objects/subjects of copyright; what is protected and what is not. ConsultantPlus+1
  • Rospatent registration of programs for computers and databases (forms and order). Russian Patent Office +1
  • USCO — the policy of registration of work with AI materials (policy statement), Report, Part 2 (January 2025) on «copyrightability,» Part 3 (May 2025, pre-pub) about learning and fair use. copyright.gov+2copyright.gov+2
  • Fair use (17 U.S.C. §107) Four factors and explanations of LII. Legal Information Institute +1
  • EU: DSM Directive (TDM, art. 3-4), and AI Act (Training transparency, disclosure pattern from 24.07.2025). Eur-Lex+2Eur-Lex+2
  • Practice and news: NYT v. OpenAI/Microsoft (procedural solutions, data retention), Ziff Davis v. OpenAI (publisher lawsuit), Getty v. Stability AI (UK: narrow solution, part of the claims rejected; TM violations confirmed). Guardian +4Reuters +4AP News +4