On May 12, 2026, Justice Minister Akın Gürlek announced the UYAP AI Karar Destek Sistemi (UYAP AI Decision Support System), a domestically built tool wired into Turkey's National Judiciary Informatics System. It can scan more than 30 million precedent decisions and complex case files in seconds to help judges and prosecutors with legal-text review, jurisprudence research, and classification of similar cases. Gürlek was emphatic that the system is "human-centered" and that the final decision always rests with the judge or prosecutor — never the machine.
On the merits of its design, this is the right call, and the editorial reflex to recoil from "AI in the courtroom" should be resisted. But design is not the same as governance, and in Turkey the gap between the two is where the real argument lives.
The tool itself is the good kind of judicial AI
UYAP is not a novelty bolted onto Turkish justice. Launched in 2000 and completed by 2007, it already connects every court, prosecutor's office, and enforcement office in the country, touching nearly 6,000 judges and 3,700 prosecutors. Adding a retrieval-and-research layer to a system that already holds the case law is the most defensible form of legal AI there is.
The distinction that matters is between decision support and decision automation. A tool that surfaces relevant precedent and flags similar cases is functionally a better search engine — it accelerates the research a clerk would otherwise do by hand. That is categorically different from "predictive justice" systems that score a defendant's recidivism risk or forecast a case outcome, which import opaque correlations into a process that is supposed to turn on reasons. UYAP AI, as described, sits firmly on the assistive side of that line.
That placement tracks the emerging global consensus almost exactly. Just three weeks after Gürlek's announcement, on June 3, 2026, the Supreme Court of India published its draft Regulations for Use of Artificial Intelligence (AI) in Courts, 2026, open for public comment until June 20. Its core principle is identical to Gürlek's framing: AI may assist courts but "can never replace judicial decision-making," and it explicitly bars AI-driven adjudication, bail-risk scoring, and outcome prediction while permitting legal research, citation verification, translation, and transcription. The Council of Europe's CEPEJ European Ethical Charter of 2018 — binding on Turkey as a Council member — said the same thing earlier, anchoring its "under user control" principle on the idea that the user must remain an informed actor and never be bound by an AI suggestion. Turkey's tool is being built to a recognised template, not invented in a vacuum.
The strongest case for the critics
Here the opposing view deserves its full weight, not a caricature. On April 27, 2026, 78 Turkish bar associations issued a joint statement opposing the expansion of AI in the judiciary, warning that "the right to defense cannot be delegated to artificial intelligence" and that such systems could erode fair-trial guarantees and public trust. This is not Luddism. It rests on a specific, well-founded anxiety about context.
The concern is automation bias. A judge formally retains the final say, but if the system pre-classifies a case as "similar" to a cluster of convictions and surfaces precedent that points one way, the path of least resistance is to follow it. "Human-in-the-loop" degrades into "human-rubber-stamp" when the human is overworked — and Gürlek himself justified the tool by pointing to judicial workload. Worse, an AI trained on 30 million past Turkish decisions will faithfully reproduce the patterns in that corpus, including any politicised jurisprudence baked into it. In a system where, as recently as 2025, the elected leadership of the Istanbul Bar Association was removed by a court order before being acquitted in 2026, the lawyers' warning that automation could entrench rather than neutralise institutional bias is serious and specific.
Proportionate regulation, not prohibition
The right response is not to block the tool — efficiency gains in a backlogged system are a genuine public good, and the bar associations notably stopped short of demanding an outright ban. The right response is to pair deployment with the guardrails the technology's own design implies but cannot self-enforce.
Three are essential, and all three already exist in the frameworks Turkey could simply adopt:
- Disclosure. India's draft would require advocates and litigants to disclose AI assistance in pleadings. The judicial mirror of that rule — recording when and how UYAP AI informed a case file — lets appellate courts and defense counsel see the machine's footprint.
- Auditability. India's draft proposes AI incident databases, annual audits, and transparency reports. Without a logged, reviewable trail, "the judge decided" is unfalsifiable.
- Contestability. The CEPEJ Charter's user-control principle means a defendant must be able to challenge an AI-surfaced classification as readily as any other piece of the record.
None of this requires slowing the rollout. It requires publishing the model's scope, its training corpus, and an audit protocol alongside it. The technology Gürlek unveiled is, on paper, the responsible kind. Whether it strengthens Turkish justice or quietly narrows it will be settled not by the quality of the code but by whether Ankara is willing to make the system as transparent as it is fast.