Copyright in the Age of Artificial Intelligence: Ownership of AI-Generated Works

Copyright is one of the oldest and most sophisticated areas of law protecting intellectual creations. Within this doctrinal framework — built upon the safeguarding of human creativity — the entry of artificial intelligence (AI) into the creative process raises questions to which traditional legal principles provide only partial answers. Regulation (EU) 2024/1689 of the European Parliament and of the Council (hereinafter: the AI Act) has been applicable since 2 August 2025 as regards its provisions on general-purpose AI models, while its requirements for high-risk systems will enter into full effect on 2 August 2026. This regulatory turning point is fundamentally reshaping expectations in copyright law as well.

The Requirements for Copyright Protection and AI Creations

The Human Author Requirement in Hungarian Law

Under Section 1(1)–(3) of Act LXXVI of 1999 on Copyright (hereinafter: the Copyright Act, Szjt.), copyright protection extends to all literary, scientific, and artistic works, provided they possess an individual, original character. Hungarian judicial practice and legal scholarship consistently hold that the concept of a copyrighted work inherently presupposes human intellectual activity: copyright protection may attach exclusively to the creative effort of a natural person.

From this doctrinal premise it follows directly that a product generated autonomously by an AI system — without substantive human intervention, creative decision-making, or creative control — does not enjoy copyright protection. This does not mean that such a work qualifies as “public domain” in the traditional sense (since the concept of public domain refers to works whose protection period has expired), but rather that no copyright protection arises ab ovo — anyone may freely use the output, but no one acquires exclusive rights over it.

AI as a Tool: The Question of “Hybrid Works”

By 2026, judicial practice and legal scholarship have developed a more nuanced framework for assessing AI-assisted creations. The decisive question is the extent to which the human exercised creative control over the creative process:

• Mere prompt provision (e.g. “create a painting of a sunset”): Such a level of instruction does not embody sufficient individual, original intellectual activity to warrant copyright protection of the result. It was the AI’s autonomous decisions that determined the composition, colour palette, and details.
• Iterative creative control (detailed, multi-step prompting; manual selection, modification, combination, and post-processing of the output): Where the creator’s activity pervades the final result to such a degree that his or her individuality is reflected therein, the work may enjoy copyright protection as a hybrid creation. In such cases, the author is the human, and the AI is merely a tool — analogous to the way a camera serves as the photographer’s instrument.

Important: The determination of authorial status is always a matter of individual assessment. There is no mechanical rule as to how many prompts or what degree of intervention gives rise to copyright protection — the court evaluates the overall picture.

Training Data and the Protection of Copyright

Text and Data Mining (TDM) Rules

The copyright dimension of AI model development is not confined to the assessment of the output — an equally weighty issue is the lawfulness of using the training data (input). Articles 3–4 of the Digital Single Market Directive (DSM Directive, 2019/790/EU) regulate text and data mining (TDM) exceptions.

Article 4 of the DSM Directive permits rights holders to expressly reserve (opt out of) the use of their works for machine learning purposes. In 2026, the AI Act’s transparency obligations [Article 53(1)(c) and (d)] reinforce this protection:

• Providers of general-purpose AI models must publish a detailed summary of the content used to train the model.
• They must demonstrate that the use did not infringe the rights holders’ right to object.
• Disregarding a rights holder’s reservation constitutes a copyright infringement, giving rise to liability for damages.

Practical significance: For visual artists, photographers, writers, and composers, it is of paramount importance to indicate their TDM objection in a machine-readable format within their works or their metadata (for example, in the website’s robots.txt file or through a notice under the Copyright Act). Failure to do so may give rise to a presumption of tacit consent.

Watermarking and Labelling Obligations

Article 50 of the AI Act imposes transparency obligations on providers and deployers of AI systems:

• Image, audio, or video content created using deep fake technology must be clearly labelled as AI-generated.
• In the case of textual content, a labelling obligation likewise applies where the content is published for the purpose of informing the public about matters of public interest.
• The content must contain a machine-readable watermark enabling subsequent identification.

Failure to comply entails multiple consequences:

1. Administrative fines: Infringement of the AI Act may result in fines of up to 3% of global annual turnover.
2. Forfeiture of copyright claims: A person who conceals AI involvement and presents himself or herself as the author engages in conduct contrary to the principle of good faith and fair dealing. The court may dismiss the claim for determination of authorial status on grounds of bad faith.
3. Civil liability for damages: Concealment may also establish liability for damages under Section 6:519 of the Ptk. (general tort liability), where third parties — for example, the commissioning party — would not have entered into the contract had they known of the AI origin.

Commercial Use of AI-Generated Works: Practical Issues

The deployment of AI-generated works in a business context raises numerous practical questions:

• Marketing materials, logos, illustrations: Where a business uses AI-generated content, it must document the origin and legal basis for its use. The enterprise must assess whether it intends to claim copyright protection over the content, and if so, whether the degree of human creative contribution supports such a claim.
• Software code: The same criteria apply to program code written by AI. Code generated purely by AI does not enjoy copyright protection, whereas code substantively reworked and supplemented by the developer does.
• Contractual arrangements: Licence agreements governing AI-generated content must expressly address the role of AI, the question of any copyright, and the allocation of liability.

Summary and Practical Recommendations

The intersection of artificial intelligence and copyright in 2026 is defined by the following principles:

1. No copyright protection without a human author. The autonomous product of AI is not protected.
2. The degree of creative control is decisive. Iterative human involvement reflecting individuality may establish copyright protection.
3. Training data use is not a lawless zone. The rights holders’ right to object must be respected, and developers must demonstrate lawful use.
4. Watermarking and labelling are mandatory. Failure to comply may result in fines, forfeiture of rights, and liability for damages.
5. Contractual arrangements are indispensable. When using AI-generated content, the legal relationships must be expressly set out in the contract.

Given the complexity of these copyright issues, we recommend seeking specialist intellectual property advice before creating, using, or commercially exploiting AI-generated content. Our office is at your disposal.