Assessing Their Compatibility with Prospective Turkish AI Legislation and Their Potential Contribution to the Turkish Legal System
Av. Asutay Duhan Meydan
Attorney at Law | Artificial Intelligence Law and Technology Policy
Italy’s Law No. 132 of 23 September 2025 does not merely reproduce the general framework established by the EU Artificial Intelligence Act. Instead, it intervenes in specific fields in which the use of artificial intelligence may create immediate legal difficulties within the domestic legal order. Its provisions concerning regulated professions, judicial activity, AI training data, copyright and deepfake content therefore offer a useful comparative model for Türkiye.
Türkiye has not yet enacted a comprehensive horizontal statute governing artificial intelligence across all sectors. Existing and proposed reforms nevertheless indicate a growing preference for risk-based regulation, institutional coordination, ethical safeguards and a legal framework that protects fundamental rights without unnecessarily restricting innovation.
Against that background, five provisions of the Italian law should not be viewed as rules to be copied verbatim. Their greater value lies in demonstrating how a national legislature may translate broad AI principles into concrete remedies, professional duties, procedural safeguards and rules of liability adapted to its own legal system.
1. A Dedicated Legal and Remedial Framework for AI Training Data
Article 16 of the Italian law authorises the Government to establish a comprehensive legal framework governing the data, algorithms and mathematical methods used in the training of artificial intelligence systems. Its most significant feature is that it goes beyond defining conditions for lawful use. It also contemplates compensation mechanisms, preventive and prohibitory remedies, sanctions and the allocation of disputes to specialised judicial divisions.
Such an approach would be highly relevant to Türkiye.
At present, the use of protected works, personal data, trade secrets or databases for model training must be assessed through several different legal regimes, including the Law on Intellectual and Artistic Works, the Personal Data Protection Law, the Turkish Commercial Code, tort law and contractual obligations. This fragmented structure makes it difficult for right holders to determine the proper cause of action, the competent court and the remedy that may realistically be granted once a model has already been trained.
A future Turkish AI statute could therefore include a dedicated chapter on training data. Such a chapter could regulate:
- documentation of data sources;
- licences and reservations of rights;
- the processing conditions applicable to personal data;
- transparency concerning training datasets;
- the removal of unlawfully obtained data;
- access and information rights of affected persons;
- and the evidential obligations of model providers.
The statute should not restrict available remedies to monetary compensation. Depending on the circumstances, courts should also be able to order the suspension of training, prohibit further use of a dataset, require deletion of unlawfully processed materials or temporarily prevent the release of a particular model version.
The Italian model of specialised business courts should not necessarily be transferred directly into Turkish law. A more suitable solution may be to allocate disputes according to their legal nature between specialised intellectual and industrial property courts and commercial courts, while establishing technical expertise and procedural coordination for AI-related claims.
The principal legal benefit would be to transform AI training from an obscure technical process governed indirectly by several unrelated legal fields into a legally identifiable activity with clearly defined rights, duties and remedies.
2. Transparency Duties for Professionals Using Artificial Intelligence
Article 13 of the Italian law provides that artificial intelligence may be used in intellectual and regulated professions only for instrumental and supportive purposes, while the professional service itself must remain centred on human intellectual activity. It further requires professionals to provide clients or service recipients with clear, plain and comprehensive information concerning the AI systems used.
A comparable rule could create considerable legal value in Türkiye, particularly in the fields of legal practice, medicine, accountancy, engineering and academic or technical consultancy.
A complete prohibition on professional use of artificial intelligence would be unrealistic. At the same time, leaving such use entirely to the discretion of the professional may generate serious risks concerning confidentiality, personal-data security, professional diligence, independence and the substantive quality of the service provided.
For example, where a lawyer uploads documents belonging to a client to an external AI system, that decision is not merely a technical matter. It may engage professional secrecy, cross-border data transfers, confidentiality obligations, the accuracy of legal analysis and liability for erroneous advice. In materially significant cases, the client’s awareness that their information or legal matter has been processed with AI assistance may form part of the professional relationship of trust.
Any Turkish disclosure obligation should nevertheless be proportionate.
It would not be necessary to disclose every incidental use of spelling correction, formatting or routine administrative tools. The duty should instead apply where:
- personal or confidential data are transferred to an AI system;
- AI is used to generate a legal, medical, financial or technical assessment;
- the system materially influences professional advice or recommendations;
- or a substantial part of the service is produced through automated assistance.
Such a provision would not prohibit AI use. Its purpose would be to confirm that human supervision, professional judgment and ultimate responsibility remain with the professional.
It would also strengthen informed consent, evidential clarity and accountability in circumstances where the client may otherwise be unaware that an external system has participated in the preparation of the professional service.
3. Preserving Human Judicial Authority
Article 15 of the Italian law expressly reserves to the judge the interpretation and application of the law, the assessment of facts and evidence, and the adoption of judicial decisions. Artificial intelligence may be used for the organisation of judicial services, administrative simplification and supporting functions, but it may not replace the exercise of judicial authority.
This principle is broadly compatible with the constitutional structure of the Turkish judiciary.
Artificial intelligence may provide substantial benefits in the classification of case files, identification of similar judgments, procedural deadline management, anonymisation, transcription, document review and the administration of court workloads. These uses may improve efficiency without interfering with the substance of adjudication.
By contrast, the assessment of witness credibility, the reliability of evidence, legal characterisation and the determination of the judgment cannot be delegated to an algorithm.
A Turkish AI statute should, however, go beyond the general statement that “the final decision belongs to the judge.” Formal human approval may be insufficient where an algorithmic recommendation becomes practically determinative.
Effective safeguards should therefore require that:
- judges understand the limitations and known error rates of the system;
- AI-generated recommendations remain reviewable and contestable;
- the use of an AI tool capable of materially influencing the proceedings is recorded;
- the parties are informed where appropriate;
- and the court remains capable of explaining its reasoning independently of the system’s output.
Without such safeguards, a judge may formally retain decision-making authority while being functionally dependent on an opaque algorithmic assessment.
The legal benefit of this model is that it allows the judiciary to use technology without undermining judicial independence, adversarial procedure or the right to a reasoned judgment. It positions AI as an auditable support mechanism rather than an invisible substitute for judicial reasoning.
4. Human Intellectual Contribution in AI-Assisted Works
Article 25 of the Italian law links copyright protection to human intellectual creation. A work produced with the assistance of artificial intelligence may qualify for protection only where it remains the result of the author’s intellectual effort.
This approach could be incorporated into Turkish copyright law without disrupting its underlying principles.
A rule excluding all AI-assisted works from protection would disregard the genuine creative effort of individuals who use AI as a tool within a broader artistic or intellectual process. Conversely, automatically granting copyright to any person who enters a short prompt would expand authorship beyond its doctrinal foundations.
The decisive criterion under Turkish law should therefore be the presence of meaningful human creative contribution.
Relevant factors could include:
- the originality of the human concept;
- the specificity and development of the instructions;
- the selection and rejection of outputs;
- repeated revisions and creative interventions;
- the arrangement and transformation of generated material;
- and the extent to which the final work reflects the personal choices of the alleged author.
The legal inquiry should not be limited to whether AI was used. The more appropriate question is whether the individual exercised genuine creative control over the final form of the work.
Such a rule would improve legal certainty for publishers, artists, software developers, designers and content creators. It would also allow contractual rights to be allocated more predictably.
At the same time, protection of the final output must remain legally distinct from the lawfulness of the training process. The fact that an AI-assisted work qualifies for copyright protection because of meaningful human input does not establish that the source materials used to train the model were lawfully obtained or processed.
A future Turkish statute should therefore separate two issues that are often incorrectly merged:
- whether the output contains sufficient human creativity to receive copyright protection; and
- whether the training data were used in compliance with copyright, data-protection and contractual rules.
5. A Specific Criminal Offence for Harmful Deepfakes
Article 26 of the Italian law criminalises the non-consensual dissemination of AI-generated or AI-manipulated images, videos or audio where the material is capable of misleading others as to its authenticity and causes unjust harm to the person concerned. The provision also treats certain uses of AI as an aggravating circumstance where the technology facilitates deception, obstructs the victim’s defence or increases the consequences of the offence.
A carefully drafted deepfake offence could fill a genuine gap in Turkish criminal law.
The Turkish Criminal Code already contains provisions concerning privacy, personal data, defamation, threats, blackmail, obscenity and offences involving children. However, a fully synthetic image or audio recording may not always fit comfortably within offences drafted on the assumption that a genuine recording, statement or event exists.
The absence of an actual original recording may create uncertainty concerning the material elements of existing offences, even where the fabricated content causes severe reputational, sexual, economic or political harm.
A Turkish deepfake offence should not, however, criminalise content merely because it was generated with artificial intelligence. The offence should require clearly defined elements, including:
- dissemination without the consent of the identifiable person concerned;
- deceptive capacity;
- intent or at least a specified degree of culpability;
- actual harm or a serious and concrete risk of harm;
- and a sufficient connection between the fabricated content and the affected individual.
The law should also include safeguards for freedom of expression, artistic works, parody, satire, public-interest journalism and fictional content that is clearly identified as such.
A specific offence would be particularly useful in cases involving:
- non-consensual synthetic sexual imagery;
- fabricated recordings intended to influence elections;
- false evidential material;
- impersonation and voice-cloning fraud;
- and synthetic content used for blackmail or reputational destruction.
Its value would not be limited to punishment. A clearly defined offence could also improve evidence-preservation procedures, facilitate police specialisation, strengthen platform reporting systems and enable victims to obtain urgent protective measures more effectively.
Conclusion
These five provisions of the Italian law should not be treated as a single legislative package that Türkiye ought to adopt without modification. They are better understood as targeted responses to distinct legal problems created by artificial intelligence.
For Türkiye, the most valuable legislative approach would not be to reproduce the general risk-based structure of the EU AI Act in isolation. It would be to identify where artificial intelligence creates uncertainty within the national legal system and to address those areas with specific rights, duties and remedies.
A dedicated regime for training data, proportionate professional transparency duties, preservation of human judicial authority, a human-contribution threshold for copyright and a narrowly defined deepfake offence could substantially improve legal certainty in Türkiye.
The objective of such legislation should not be to punish technology. It should be to make responsibility visible and legally enforceable.
A well-designed Turkish AI law should:
- protect individuals without suppressing innovation;
- preserve professional responsibility while allowing technological efficiency;
- distinguish genuine human creativity from fully automated production;
- ensure that AI supports rather than replaces judicial decision-making;
- and provide effective remedies when artificial intelligence is used to infringe dignity, privacy, property or other protected rights.
The principal lesson offered by the Italian example may therefore be expressed as follows:
An effective artificial intelligence law derives its value not from defining artificial intelligence at excessive length, but from clearly determining who is responsible, to what extent and through which legal remedy when data, creative works, professional duties, judicial decisions or human dignity are adversely affected.