- within Real Estate and Construction and Environment topic(s)
In Kazakhstan the Law "On artificial intelligence" (hereinafter the "AI Law") entered into force on 18 January 2026.
The AI Law addresses a wide range of issues, including risk management, transparency, and safety, as well as the creation of a national platform and data libraries. Among its provisions, those related to intellectual property, particularly copyright, represent some of the most anticipated and legally significant developments. For the first time in a dedicated legislative act, the state has expressly defined its approach to works created with the involvement of artificial intelligence. Artificial intelligence is not recognised as an author, and specific criteria for the protection of such works have been introduced. Additionally, special rules regulating the training of models on protected content have been established.
Article 23 of the AI Law plays a pivotal role in the context of copyright law. It establishes a fundamental principle that works created using artificial intelligence systems are eligible for copyright protection only where a human creative contribution is present. In doing so, the legislature has definitively endorsed a model under which artificial intelligence is treated as a tool rather than a subject of creative activity. This marks a shift from the theoretical question of whether such works may be protected to a practical assessment of what the human author has actually done and how their creative choices are expressed.
In practice, this approach shifts the focus from the mere fact of using artificial intelligence to an analysis of the creative process leading to the final result. Particular importance will be attached to editorial revision, composition, narrative structure, artistic and stylistic decisions, and the internal logic of the text or image. For Kazakhstani courts and for the Republican State Enterprise "National Institute of Intellectual Property" under the Committee for Intellectual Property Rights of the Ministry of Justice of the Republic of Kazakhstan (hereinafter referred to as the "Expert organisation"), the decisive factor in matters of deposit and registration of copyright objects will be the presence of a human creative contribution.
The Digital Code of the Republic of Kazakhstan, scheduled to enter into force in July 2026, further reinforce this approach at the higher legislative level. Article 23 of the Digital Code defines software as a digital object consisting of a set of codes that implement algorithms and enable the processing and transmission of data. At the same time, it expressly provides that rights to software are protected in accordance with the Republic of Kazakhstan's legislation on copyright and related rights.
Thus, from the perspective of private law, artificial intelligence systems are classified as a software, while the legal regime governing their outputs continues to be grounded in traditional copyright law. This confirms the model chosen by the legislature, under which artificial intelligence is treated as a technical tool rather than as an independent subject of intellectual property.
Courts will likely see cases in which the parties must demonstrate not merely that content was generated using artificial intelligence, but the extent and nature of the human creative contribution. Evidence may include file versions, editing histories, the structure of text prompts, drafts, scenario plans, correspondence with designers or developers, and other materials showing that the final output results from human intellectual choices rather than fully automated generation.
Of particular significance in this context is the Law's provision recognising textual prompts as copyrightable objects, provided that they themselves are the result of human intellectual and creative activity. This effectively legitimises a practice long established in creative and technological projects, where value is generated not so much at the level of the model's output, but through the design of the task itself, involving complex prompt structures, iterative processes, and carefully planned orchestration of the final result.
It is important to note, however, that the protection of a prompt does not automatically extend to the resulting output. A prompt may be protected as an independent text if it meets the standard of originality. At the same time, the final product will be assessed separately, based on the extent to which a human contributed to its creation and refinement.
This development will inevitably impact contractual practice. Agreements with designers, marketing agencies, developers, and content producers will need to address rights in prompts separately from rights in the generated outputs. In practice, new contractual structures are likely to emerge, distinguishing between rights in the original prompts, rights in the generated results, and liability for potential infringement of third-party rights.
Notably, these approaches have already begun to take shape in the Expert organisation's practice. On 31 December 2025, the National Institute of Intellectual Property (NIIP) issued a copyright certificate for the graphic novel Aisha vs. Albasty, created with the assistance of ChatGPT. Following an initial refusal in 2023, the Expert organization revised the application so that protection was sought for the work as a composite, with a natural person recognised as the author of its composition and structure. This case provides practical confirmation that the decisive factor is the human creative contribution, rather than the mere use of artificial intelligence.
Equally significant is the set of provisions addressing the training of artificial intelligence models on protected works. For the first time, the Law explicitly distinguishes between model training and the free use of works for educational or scientific purposes. Training cannot be automatically justified under such exceptions. At the same time, the Law clarifies that using works for training does not imply their use in ways that constitute the author's personal or economic rights, such as reproduction, distribution, adaptation, or public display.
In judicial practice, this implies the emergence of a new category of disputes over the lawful use of content in datasets for model training. Parties will need to consider not only whether consent or a license was obtained, but also whether a machine-readable restriction was implemented. Additionally, they should evaluate whether the developer was aware of such a restriction and whether the developer was obliged to take it into account. These factors will form the basis for disputes between rights holders, IT companies, and platform owners.
At the same time, a mechanism for opting out through a machine-readable format is introduced. Training is permitted only in the absence of a prohibition issued by the author or rights holder in such a format. This establishes a special regime under which use is allowed until a prohibition is declared, provided it is clearly communicated through a technological means. It is in this area that judicial practice is likely to develop in the coming years, as, in practical engineering terms, model training involves copying and transforming data, while the boundary between the training process and reproduction as a legal fact remains a matter of debate.
The provisions on transparency and labelling of synthetic outputs also carry practical significance. The Law requires that users be informed when goods, works, or services are produced with the use of artificial intelligence, and the distribution of synthetic content is permitted only if it is labelled in a machine-readable format and accompanied by a visible warning. In copyright disputes, this may serve as evidence to establish the origin of content and to distinguish between human creativity and machine-generated output.
The introduction of the concepts of data libraries and data library producers, along with the requirement to indicate their origin in a machine-readable format, establishes a foundation for a future infrastructure that licenses content for artificial intelligence training. In the long term, this could lead to the development of a market for legal datasets and standardised terms regarding the use of works in training.
Taken as a whole, the AI Law shifts the discussion of copyright in the age of algorithms from a theoretical plane to the realm of judicial practice and contractual drafting. For the courts, key issues will include the boundary between human creative contribution and automated generation, as well as the permissibility of using works for model training. For contractual practice, the Law signals the need to adapt agreements to the new reality, where, alongside rights in content, there are rights in prompts, rules governing model training, and obligations to label synthetic outputs.
In this way, Kazakhstan is laying the foundation for a new enforcement practice in the field of copyright, in which artificial intelligence is recognised as a tool of creativity, while legal responsibility and rights to the resulting work remain with the human author.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
[View Source]