Updated 2 November 2026: the Court already has AI tools of its own, which the earlier text overlooked. Article 11 of Law No. 6698 does not reach judicial processing, because Article 28 excludes it. The smart-cities action names areas such as traffic, energy, water, waste and security, not licences or social assistance.

Correction (19 October 2026): Türkiye has 1,402 municipalities, not about six hundred; under the published strategy document, AI pilot projects in selected cities are scheduled for 2027–2029, not 2026–2027; and the description of the Türkiye AI Action Plan (2026–2030) now follows its published actions.

On 23 September 2026 Kadir Özkaya, President of Türkiye's Constitutional Court, said that artificial intelligence must support judges and not replace them. He spoke in Ankara at the symposium marking 14 years of individual applications to the Court. He also described AI tools the Court is already building. This was one section of a speech, not a judgment, and it binds no court.

What did the Constitutional Court's President say about AI?

He said that "artificial intelligence must be used not to take the judge's place but to support the judge and the judicial decision-making process". That is from the text published by the Court. His other points were these:

  • Responsibility. Responsibility for the judicial decision ultimately belongs to a human "and must remain so".
  • Principles. If AI is used, there must be no compromise on transparency, explainability, personal data protection, prevention of discrimination and continuous human oversight.
  • Necessity. Given the nature of the system and the number of applications, using technology is "no longer just a choice but a necessity".
  • Human judgement. Concepts such as human dignity and proportionality cannot be explained by a mathematical formula based only on data.

The figures he gave: 756,726 individual applications since 23 September 2012, of which 648,410 were concluded, about 85.7%. By subtraction, that leaves about 108,000 unresolved. He said 99.9% of the 86,849 violation rulings since 2013 have been implemented. He also cautioned against seeing these numbers purely as a workload question, because each one is a person seeking justice.

Is the Constitutional Court already using AI?

Yes, for preparatory work, according to the President. An earlier version of this article described these systems as mostly not yet existing, which was wrong.

  • In-house systems analyse and summarise application forms and related court files, to help rapporteurs.
  • Scanned paper applications are converted into text.
  • The models run on a closed-circuit infrastructure with no internet connection, for data security.
  • With TÜBİTAK, a decision-support system for admissibility review is "nearing completion", to help rapporteurs find related applications and case law.

In April he said the outputs would have no legal value and be used only at the preparation stage, according to a press report. On 2 October he told a congress of Asian constitutional courts that a structured application file summary was now in place, and that AI "cannot become an authority over the judge".

Which legal rules stand behind these principles?

Each has a source in Turkish law, although the speech did not cite them. This mapping is our reading.

  • Human responsibility. Articles 9 and 138 of the Constitution give judicial power to independent and impartial courts, with judges deciding according to their conscience.
  • Transparency and explainability. Article 141 requires every court decision to be written with reasons. A contribution to an outcome that cannot be explained is hard to reconcile with that.
  • Non-discrimination. Article 10 of the Constitution and Article 14 of the European Convention on Human Rights apply. Systems trained on past outcomes can repeat the patterns in them.
  • Personal data. Article 20 of the Constitution protects personal data. Law No. 6698 needs care here: Article 28(1)(d) excludes processing by judicial authorities in connection with investigation, prosecution, trial or enforcement. So the Article 11 right to object to automated results, which an earlier version applied to judicial outcomes, does not simply carry over.
  • Human oversight. Review that exists only on paper is not oversight. In our view, reviewers under heavy workload are where this risk is highest.

What does the smart-cities circular require of municipalities?

It sets priorities rather than ordering any specific AI decision. Presidential Circular 2026/11, published in the Official Gazette of 19 September 2026, puts the 2030 National Smart Cities Strategy and Action Plan into effect. The circular says priority will be given to "AI-supported decision-making mechanisms", the integration of smart-city services and digital city management.

The strategy document has 32 actions. Action 9 aims to spread AI in smart-city services. It names traffic, energy, waste, water and security, as well as automating city-management processes and making services more personalised. Priority areas and solutions are to be chosen first, with pilots in selected cities and an evaluation of large language models planned for 2027–2029. The indicator is 30 municipalities using AI by 2027, rising to 250 by 2030.

An earlier version said municipalities were being told to automate decisions such as licences and social assistance. The action does not say that. Those are the cases where the legal questions would be sharpest if AI spread to them, and that is the reason to discuss them.

What does administrative law already require?

Existing rules would apply to any AI-supported municipal decision. These points are our reading.

  • Judicial review. Under Article 125 of the Constitution, all acts of the administration are open to judicial review. Article 2 of Law No. 2577 lets courts examine authority, form, reason, subject and purpose. A score alone does not explain the reason for an act.
  • Authority. Power belongs to the organ the law names. In our view, an official who merely signs what a system produces raises a question on that element.
  • Automated results. Article 11(1)(g) of Law No. 6698 lets a person object to an adverse result produced exclusively by automated analysis. Municipalities are not within the judicial exclusion in Article 28(1)(d). Article 28(2)(c), however, lifts most of Article 11 where a public body needs the processing for statutory inspection, regulation or discipline, so the right will not cover every municipal decision. "Exclusively" has not, as far as we know, been tested in this setting.
  • Records and procurement. Review needs a record: inputs, model version, output, the official and the reasons. Audit rights and retention are best settled in the contract.

Is there a gap between the principles and practice?

We found no binding Turkish rule specific to AI in public decisions as of 1 November 2026, but this may change. The AI Action Plan (2026–2030), under Circular 2026/9, says in Action 7 that guides, checklists and templates will be published for public bodies, and that high-impact systems will go through certification and evaluation. KVKK has published guidance on AI, including a paper on agents. Neither the speech nor the circular refers to the other, so comparing them is this article's framing.

In the EU, Annex III of the AI Act treats two such uses as high-risk: systems used by public authorities to assess eligibility for essential public benefits and services (point 5(a)), and systems that help judicial authorities research and interpret facts and law (point 8(a)). These duties apply from 2 December 2027 under Regulation (EU) 2026/1744. Türkiye has no equivalent in force. Our high-risk assessment guide explains the test.

What should a municipality consider before a pilot?

  • Classify the use first. Is the output advisory or decisive, and does it affect a right or obligation? Traffic signal timing and a benefit refusal are different cases.
  • Keep the decision record. Without it, a court cannot review the act and the municipality cannot defend it.
  • Write reasons from the file, not the score. The reasoning should stand on the underlying facts.
  • Measure departures. In our view, if officials never disagree with the system, oversight deserves a closer look.
  • Fix audit rights at procurement. Model documentation, retention and notice of changes are hard to add later.

Frequently asked questions

Did the Constitutional Court rule on artificial intelligence?

No. These were remarks by the Constitutional Court's President at a symposium on 23 September 2026. They have no binding force. As far as we know, the Court has not ruled on AI in adjudication, so how it would decide such a case remains open.

Does the Turkish Constitutional Court use AI to decide cases?

According to its President, no. The Court's in-house systems summarise and analyse files for rapporteurs on a closed network, and a decision-support tool for admissibility review is nearly complete. He said responsibility for the decision remains human.

Does Circular 2026/11 require municipalities to use AI?

Circular 2026/11 gives priority to AI-supported decision-making and puts a strategy into effect. Action 9 of the strategy sets a target of 250 municipalities using AI in smart-city applications by 2030. It is a policy instrument, and we found no rule requiring a particular municipality to use a particular system.

Does KVKK Article 11 apply to courts and municipalities?

The two differ. Article 28(1)(d) of Law No. 6698 excludes processing by judicial authorities connected with trial and enforcement. Municipalities are public controllers, so in our reading a person can generally rely on Article 11(1)(g) for an adverse result produced exclusively by automated analysis, subject to the Article 28(2) exceptions for statutory inspection, regulation and discipline.

Burhan Doğuş Ayparlar's View

This section sets out my personal assessment as the founder of this site and an attorney at law (Türkiye).

In my view, the most useful part of this speech is what sits alongside the principles: a real court describing real tools on a closed network. It is also where the hard question lies. The case for the technology rests partly on the number of applications. The safeguard is a human who can disagree, and that takes time. The pressure that justifies the tool also presses on the safeguard. This has to be resolved by design: limiting what the system touches, and measuring how often reviewers depart from it.

The municipal side concerns me more. Most of the strategy's named examples are services such as traffic and waste, which raise fewer rights questions, although it also names security. But if AI moves towards decisions about individuals, I would start with a record-keeping rule rather than a statute. Any administrative decision in which an AI system materially contributed should keep the inputs, the model version, the output, the deciding official, and reasons drawn from the file. That preserves judicial review, which is what protects the citizen, and it costs little if specified before procurement.

Sources: text of the remarks of 23 September 2026 and 2 October 2026 published by the Constitutional Court; Circular 2026/11 (Official Gazette No. 33375, 19 September 2026); 2030 National Smart Cities Strategy and Action Plan of the Ministry of Environment, Urbanisation and Climate Change (published at akillisehirler.csb.gov.tr; that host did not respond on 1 November 2026, so we relied on a copy read on 19 October 2026); AI Action Plan, Action 7; Constitution; Laws Nos. 6698 and 2577; Regulations (EU) 2024/1689 and 2026/1744; Memurlar.Net (29 April 2026, press report). General information, not legal advice.