Turkey's justice ministry has deployed an AI tool that scans criminal case files for links to groups the government designates as terrorist organizations, a system human rights lawyers say risks automating flawed legal assumptions in a judiciary already criticized by Europe's top human rights court.
The Chief Public Prosecutor's Office Organization Prediction Project was built by the ministry's Directorate General of Information Technology. It analyzes data in Turkey's National Judiciary Informatics System (UYAP) and compares it against records of groups Ankara classifies as terrorist organizations. The ministry says the tool corrects data entry errors, improves judicial statistics, and helps prosecutors and judges reach decisions.
It is a form of AI for Legal - machine learning trained on court records to flag potential organizational links. The tool is one of nine ministry AI projects, and by June 2026 it was available to all judges and prosecutors nationwide.
Lawyer Hatice Yıldız questioned the system on X this week. "A project to predict the [terrorist] organization has been developed. The prosecutor's office will upload the information in the investigation file and AI will identify the organization, supposedly preventing human error," she wrote. "Does the prosecutor not understand the files they read? What is this, some kind of [terrorist] affiliation detector?"
How the tool works
Justice Ministry Information Technology Director General Servet Gül presented the system in April 2025 as a way to fix data entry mistakes in prosecutor files. He said it automatically matched new entries with existing terrorist organization records and would improve reporting requested by international bodies such as the Financial Action Task Force. The ministry's June 2026 description says the system analyzes a case file's contents, predicts a possible organization link, and assists judges and prosecutors in decision making.
Ali Yıldız, a human rights lawyer based in Belgium, said the system could reproduce legal assumptions embedded in past terrorism investigations and court rulings. "Machine-learning systems derive their predictions from historical data," he said. "Where that data reflects legal interpretations that the European Court of Human Rights [ECtHR] has already held to be incompatible with the European Convention on Human Rights, the resulting model risks perpetuating - and potentially reinforcing - those very interpretations."
Rights concerns
The ECtHR's Grand Chamber ruled in Yüksel Yalçınkaya that Turkish courts created an almost automatic presumption of guilt by treating use of the ByLock messaging app as proof of terrorist organization membership. In its May 2026 judgment in Şaban Yasak, the court found Turkish courts failed to assess through concrete evidence whether the defendant knew of the organization's aims and intended to take part in them.
"A system trained on years of Turkish terrorism investigations and prosecutions, for example, may replicate the same generalized indicators and presumptions of organizational membership that the Grand Chamber condemned in Yalçınkaya and Yasak while cloaking those flawed assumptions in the appearance of objective, algorithmic neutrality," Yıldız said.
The Council of Europe's European Commission for the Efficiency of Justice says AI used in judicial systems should respect fundamental rights, prevent discrimination, use secure and verified data, permit outside audits, and remain under the control of informed users. Yıldız said the ministry has disclosed almost nothing about the system's training data, performance metrics, false-positive rate, or safeguards to detect bias, and has not said whether defendants will be told if an AI assessment influenced their case.
The deployment follows a decade of upheaval in Turkey's judiciary. After a 2013 corruption investigation targeted senior government figures, President Recep Tayyip Erdoğan accused the prosecutors and judges involved of acting for the Gülen movement. Following the July 2016 coup attempt, the government dismissed more than 4,000 judges and prosecutors over alleged Gülen ties. The movement denies involvement. Critics say the purge created a chilling effect that pushed remaining judges toward government priorities.
Why this matters for government professionals
The Turkish case is a warning about deploying AI in high-stakes decisions without transparency. Public agencies considering similar tools for legal document review and compliance checks should demand answers about training data, error rates, and audit trails before rollout. The same scrutiny applies to AI for Government projects more broadly: an algorithm that quietly encodes past mistakes is harder to fix than the human errors it replaces.
Before approving any AI system, government teams should ask three questions: What data trained the model, what is its error rate, and who reviews its output? The Turkish system raises each of these without clear answers. Agencies that skip those checks risk building tools that look objective but embed the biases of the records they were trained on, eroding public trust in the process they were meant to improve.
Your membership also unlocks:
