Artificial Intelligence and Intellectual Property: Who Owns the Result When the “Author” Is a Machine?
Content of the article
Generative artificial intelligence is no longer a technological novelty — it has become an everyday tool for lawyers, designers, developers, and businesses. However, Ukrainian and international law have not yet provided comprehensive answers to a fundamental question: if a work, code, or image is created by a machine, who owns it? This article systematizes the existing legal framework, analyzes key court precedents, and provides practical recommendations for businesses.
The Problem of Legal Personality: AI as a Tool, Not an Author
What current legislation says
The legal systems of Ukraine, the European Union, and the United States follow a shared core principle regarding authorship: only a natural person can be an author. This is not a technical rule but a conceptual foundation of copyright law.
In Ukraine, Article 1 of the Law “On Copyright and Related Rights” defines an author as a natural person whose creative work produced the work. Legal entities and technical tools cannot be recognized as authors — they may only acquire economic rights to a work.
In EU law, the DSM Copyright Directive (2019/790) and the AI Act (2024) do not grant legal personality to AI systems. The AI Act classifies AI models and systems as products or services — without any attributes of a legal subject.
In the United States, the U.S. Copyright Office clarified in 2023–2024 that works created solely by AI without “sufficient human creative control” are not eligible for copyright protection. The Thaler v. Perlmutter (2023) decision confirmed that courts refuse copyright registration for works where an AI system is listed as the sole “author.”
AI as a tool: the camera analogy
A convenient analogy for understanding AI’s legal status is a camera. The camera captures the image, but copyright belongs to the photographer who made creative decisions about angle, composition, and timing. The same applies to AI: the system generates content, but the legal subject is the human who formulates the task, defines parameters, and makes creative decisions.
The degree of human creative contribution is critically important. The more detailed and specific the prompt, and the more a person edits and directs the result, the higher the chance that copyright will be recognized. Conversely, if someone simply clicks “generate” without creative input, legal protection becomes highly questionable.
Objects of “Non-Obvious” Authorship: Who Owns AI-Generated Content
2.1. Three potential rights holders
When an AI system generates code, design, text, or images, the question arises as to how rights are distributed among three categories of actors:
- The AI model developer (e.g., OpenAI, Google, Anthropic, Stability AI)
- The operator — a company or entrepreneur integrating AI into their product or service
- The end user who creates prompts and uses the results
In practice, the answer to the allocation of rights is usually found in the Terms of Service (ToS) of a specific AI platform rather than in general copyright law.
2.2. What Terms of Service usually define
Analysis of ToS of leading AI platforms shows a common trend: most companies waive claims to authorship of generated outputs and transfer rights to the user. For example:
- OpenAI (ChatGPT, DALL-E): under current terms, output is assigned to the user to the extent OpenAI has rights in it.
- Midjourney: free plan — Creative Commons NC 4.0; paid plans — commercial rights are transferred to the user.
- GitHub Copilot: Microsoft explicitly states it does not claim authorship of generated code; compliance responsibility rests with the user.
Important caveat: the transfer of rights from the platform to the user is only a contractual arrangement and does not resolve the issue of copyright protection. If the material does not meet the threshold of “sufficient human creative contribution,” it may not be protected by copyright at all, regardless of what the ToS states.
2.3. Practical scenarios and recommendations
Scenario A. AI-generated code
A developer uses GitHub Copilot or ChatGPT to write code. If the developer actively contributes creatively — defining architecture, algorithms, editing, and integrating the result — copyright arises in their favor. If the code is copied “as is” without meaningful adaptation, legal protection remains questionable. An additional complication: Copilot may have learned from third-party licensed code (see Section 3).
Scenario B. AI-generated design and images
A marketer creates a logo using Midjourney. The strength of legal protection depends on the complexity and specificity of the prompt, the number of iterations, and the degree of human editing of the final image. Registering a trademark for a purely AI-generated logo without sufficient human creative input will be vulnerable.
Scenario C. AI-generated legal documents
A company uses AI to prepare contract templates. The author of the final document will be the lawyer or the company if they substantially edited, structured, and adapted the output. If the document is simply a “rephrased prompt,” authorship remains uncertain. In any case, legal responsibility for the document always lies with a human.
Risks of Using Open Source Data: How AI Can Infringe Your Copyright
3.1. How AI models are trained and where the problem arises
Modern large language models and diffusion-based image generators are trained on massive datasets from the internet containing billions of texts, images, code fragments, and other materials. A significant portion of these materials is protected by copyright or licensed under terms that restrict commercial use.
The problem arises at two levels:
- Model developer level: Did the developer have the right to use protected materials for training?
- End-user level: Does the generated output contain protected elements from the training data?
3.2. Landmark court cases
GitHub Copilot — Doe v. GitHub (2022–2024)
A class of software developers filed a lawsuit against GitHub, Microsoft, and OpenAI, alleging that Copilot was trained on their code hosted on GitHub under open-source licenses (MIT, GPL, etc.) without complying with the terms of those licenses. In particular, Copilot allegedly reproduced code fragments without attribution and without complying with copyleft license requirements.
Key legal argument: an open-source license does not mean the absence of copyright. MIT or Apache 2.0 is not “public domain,” but a license with specific usage conditions. Violating those conditions constitutes copyright infringement.
Stability AI, Midjourney, DeviantArt — artists’ lawsuits (2023)
Groups of artists in the US and UK filed lawsuits against companies developing image generators, alleging direct copyright infringement by training on billions of images without permission or compensation. In parallel, Getty Images filed a lawsuit against Stability AI for the unauthorized use of 12 million photographs from its database.
The New York Times v. OpenAI and Microsoft (2023)
The New York Times filed a lawsuit claiming that ChatGPT and Copilot were trained on millions of its articles without permission and can reproduce their content almost verbatim. The case raised concerns about competitive displacement: AI systems can provide answers based on content that publishers previously monetized through subscriptions.
3.3. The concept of “memorization” and business risks
Research shows that large language models can reproduce — fully or partially — fragments from their training data. This phenomenon is known as memorization. The practical risk is that a company may unintentionally publish copyrighted text or code in its product without even realizing it.
A particularly serious risk arises when AI-generated code is used in commercial products without verifying its origin. If Copilot reproduces GPL-licensed code in a proprietary product, the legal consequences may include an obligation to disclose the entire source code of the product (the “viral” effect of GPL).
Recommendations for Business
4.1. Legal hygiene when using AI tools
Carefully review the Terms of Service of AI platforms before commercial use. The terms change regularly — appoint a responsible person to monitor updates.
Document the creative process. Keep prompts, iterations, versions, and records of human edits — this is your evidence of a “sufficient creative contribution” in case of a legal dispute.
Separate AI-generated and human-created content in internal systems. For mission-critical materials (logos, patent applications, key contracts), do not rely solely on AI.
Do not register copyright in purely AI-generated materials — such registration will be vulnerable and may be invalidated.
4.2. Safe use of AI in software development
Introduce an internal AI Code Review policy: all AI-generated code must be checked for similarity to known licensed sources.
Use specialized tools to scan code for license obligations (e.g., FOSSA, Black Duck, ScanCode).
For closed commercial products, choose AI tools that offer commercial guarantees and indemnification (e.g., GitHub Copilot Business, Amazon CodeWhisperer for Business).
Maintain a registry of AI tools: which systems are used, for what purposes, and under what terms.
4.3. Protecting your rights to AI content
Develop an internal company AI policy regulating: permitted tools, documentation procedures, ownership of results, and restrictions on confidential data.
In agreements with contractors and freelancers who use AI for your projects, explicitly regulate the transfer of rights to AI-generated content.
For maximum brand protection, combine AI generation with substantial human creative input and document this process.
Pay close attention to confidentiality: do not input trade secrets, client personal data, or NDA-protected information into public AI platforms.
4.4. Contractual regulation
We recommend including the following provisions in corporate agreements, particularly IT services agreements, license agreements, and software development contracts:
AI disclosure: whether AI was used to create deliverables and, if so, how.
Originality warranty: the contractor guarantees that AI-generated content does not infringe third-party rights.
Indemnification: allocation of liability in case of third-party copyright claims.
License audit: the customer’s right to verify compliance of AI tools with licensing requirements.
Share
Frequently Asked Questions about AI and Intellectual Property . Below are answers to the most common questions received by our practice from clients working with artificial intelligence tools.
Can AI be officially registered as an author or inventor?
If I paid for a ChatGPT or Midjourney subscription — do I own the generated content?
Can a trademark be registered for a logo created by AI?
Am I legally responsible if an AI tool infringes someone’s copyright?
Is it legal to train your own AI model on third-party texts, images, or code?
What is “text and data mining” (TDM) and how is it regulated in the EU?
How can I protect my knowledge base or dataset from being used to train other AI models?
What is the “viral effect” of GPL and why is it dangerous when using AI-generated code?
How does the European Union AI Act regulate these issues?
Do I need to disclose that content was created using AI?
Can AI be officially registered as an author or inventor?
No. As of 2026, no legal system in the world recognizes AI as a subject of copyright or patent law. In Thaler v. Vidal (USA, 2022) and parallel cases in the EU and the UK, courts unanimously refused to register patents where an AI system (DABUS) was listed as the inventor. Similarly, the U.S. Copyright Office refuses to register copyright for works where the only listed author is AI. Only a natural person can be an author or inventor.
If I paid for a ChatGPT or Midjourney subscription — do I own the generated content?
In general — yes, but with important caveats. Most commercial AI platforms transfer proprietary rights to the output to the user (according to their Terms of Service). However, this is a contractual transfer of rights from the platform, not the emergence of copyright in the classical sense. If your content lacks sufficient human creative input, it may not be protected by copyright at all — meaning anyone could freely copy it. For commercially critical materials (logos, advertising campaigns, product design), it is recommended to combine AI generation with substantial human editing and document the process.
Can a trademark be registered for a logo created by AI?
Trademark registration and copyright are different legal regimes. A trademark is not registered based on creativity, but on its ability to distinguish goods or services of one entity from another. Therefore, an AI-generated logo can technically be registered as a trademark (if it meets distinctiveness requirements and does not conflict with existing marks). However, copyright protection for such a logo will be weak or absent. Recommendation: for reliable brand protection, combine trademark registration with significant human design refinement.
Am I legally responsible if an AI tool infringes someone’s copyright?
This is one of the key unresolved issues currently being litigated in the U.S. and the EU. The current trend in case law suggests responsibility may be shared between the AI model developer (for training-related actions) and the end user (for public use of generated content). If you knowingly use AI-generated content containing protected materials and publish it commercially, the risk of liability is real. Some platforms (e.g., GitHub Copilot Business, Google Workspace) offer commercial indemnification and assume part of the risk. Carefully review their terms.
Is it legal to train your own AI model on third-party texts, images, or code?
There is no clear answer even in leading jurisdictions. In the U.S., courts are assessing whether model training falls under the fair use doctrine — with inconsistent outcomes so far. In the EU, the DSM Directive includes an exception for text and data mining (TDM), but only for non-commercial research or where rights holders have not explicitly opted out. If you plan to train your own model on external data, a detailed legal analysis of sources, licenses, and jurisdiction is required before starting the project.
What is “text and data mining” (TDM) and how is it regulated in the EU?
Text and Data Mining (TDM) is the automated analysis of large volumes of digital materials to identify patterns, which forms the basis of AI training. EU Directive 2019/790 (DSM) establishes two levels of permitted TDM: - for scientific research by non-commercial organizations (broad exception); - a general exception for any entities, but rights holders may opt out via machine-readable mechanisms (e.g., robots.txt or metadata). Commercial TDM without opt-out is lawful only if licensed or if the content is publicly available without restrictions.
How can I protect my knowledge base or dataset from being used to train other AI models?
There are several layers of protection. First, legal: include clear TDM prohibitions in your website or database Terms of Use and use machine-readable opt-out mechanisms (X-Robots-Tag, robots.txt, metadata). Second, contractual: explicitly prohibit AI training use in agreements with partners and clients. Third, technical: restrict access via authentication, API limits, CAPTCHA, and traffic monitoring. Fourth, database rights: in the EU (and potentially Ukraine under EU integration), original databases are protected by a sui generis right independent of copyright.
What is the “viral effect” of GPL and why is it dangerous when using AI-generated code?
The GNU General Public License (GPL), one of the most common open-source licenses, includes a copyleft clause: if you include GPL code in your product, the entire product (including your original code) must also be released under GPL. This is the “viral effect.” AI risk: if GitHub Copilot or another AI tool reproduces GPL-licensed code in your proprietary commercial product, you may be required to open-source the entire codebase. Protection: scan AI-generated code with specialized tools before use and prefer enterprise AI tools with licensing guarantees.
How does the European Union AI Act regulate these issues?
The EU AI Act (Regulation 2024/1689) — the first comprehensive AI law — does not directly resolve authorship issues but introduces important related requirements. Providers of GPAI models (general-purpose AI such as GPT-4, Claude, Gemini) must: disclose summaries of training data respecting copyright law; comply with EU copyright rules during training; and for systemic-risk models, conduct additional assessments. The AI Act will be phased in from 2024–2027 and applies to companies offering or using AI services in the EU market, including Ukrainian companies working with EU clients.
Do I need to disclose that content was created using AI?
There is currently no general global requirement to disclose AI-generated content in most jurisdictions, including Ukraine. However, key exceptions exist: the EU AI Act requires labeling of deepfake and election-related AI content. Platforms (Facebook, YouTube, Google) impose their own disclosure rules. In advertising, transparency requirements may arise under consumer protection laws. In B2B relationships, disclosure may be contractually required. Recommendation: even where not mandatory, transparency is a sign of responsible business practice and reduces reputational risk.
We use cookies to improve the performance of the site and enhance your user experience.
More information can be found in our Privacy Notice