Friday, 11 September 2026
Artificial intelligence is beginning to enter one of the areas where automation raises some of the most difficult questions about trust, accountability and human judgement: the justice system. Azerbaijan is developing a model in which AI assists citizens in understanding legislation and helps judges organise increasingly large volumes of legal information, while formally keeping judicial decisions and responsibility in human hands.
The approach was presented at AI4 2026 by the National Artificial Intelligence Center of the Republic of Azerbaijan, which has been developing legal AI tools alongside the country’s wider national artificial intelligence programme. Azerbaijan approved its Artificial Intelligence Strategy for 2025–2028 in March 2025, establishing a national framework covering skills, governance, infrastructure, research and public-sector adoption. The government has also explicitly included information-security assessment and responsible implementation among the requirements for AI used by state institutions.
One of the first visible results is E-Qanun.ai, an AI-based platform built on Azerbaijan’s unified legislative database. The system was officially presented in September 2025 after being developed by the National Artificial Intelligence Center in cooperation with the Ministry of Digital Development and Transport and the Ministry of Justice. Its purpose is to allow users to search and analyse legislation using ordinary language rather than relying exclusively on traditional legal-document searches.
The underlying problem is straightforward. Making legislation publicly available does not necessarily make it understandable. Citizens may have access to thousands of laws, regulations and amendments but still struggle to determine which provisions apply to an employment dispute, contract, administrative procedure or other everyday legal problem. Traditional search systems are often designed around document names, legal terminology and keywords, while citizens generally approach the law through practical questions about what they are allowed to do, what rights they have and what action they should take.
Azerbaijan’s approach is therefore to place a conversational layer over the legal database. Instead of requiring a user to know the correct article number or legal phrase, the system attempts to understand the question, locate relevant legislation by meaning and produce an explanation linked back to the official legal material. The important design principle is that the answer should remain verifiable. The AI is intended to guide the user towards the law rather than become an independent source of legal authority.
That distinction is particularly important in legal applications because a confident but incorrect answer could have serious consequences. The National Artificial Intelligence Center said during the AI4 presentation that the platform has indexed tens of thousands of legal documents and millions of individual sentences, while monitoring changes to legislation so that the underlying information remains current. The organisation also presented very high internal accuracy and citation-performance figures. These should be understood as performance metrics reported by the project team rather than independently audited measures of legal correctness.
The citizen-facing platform is only the first stage of the programme. Azerbaijan is also preparing an AI system intended to assist judges with case preparation. The proposed E-Court AI platform is designed to review case files, identify relevant facts and arguments, locate applicable legislation and precedents and prepare a structured draft that a judge can examine, amend or reject.
The distinction between assistance and decision-making was repeatedly emphasised during the presentation. The system is not intended to function as an autonomous judge. Judicial responsibility remains with the human judge, who continues to interpret the evidence, apply the law and issue the final ruling.
That design reflects the wider challenge facing governments adopting AI in justice systems. A legal process involves far more than retrieving information. Judges assess credibility, context, competing arguments and individual circumstances while operating within constitutional and procedural safeguards. Automating information preparation may be feasible much earlier than automating the judgement itself.
The potential efficiency gains nevertheless explain why judicial authorities are interested. Azerbaijan has a relatively small number of professional judges in proportion to its population. Council of Europe data for 2024 recorded approximately 6.2 judges per 100,000 inhabitants in Azerbaijan, compared with a 2023 EU median of 21.5. Azerbaijan’s judiciary has expanded in recent years, but the country remains well below the European benchmark in judicial personnel per capita.
This makes administrative productivity particularly relevant. A large portion of a judge’s working day can be consumed before substantive judicial reasoning begins: reading case files, identifying relevant claims, comparing previous decisions, locating applicable legislation and organising supporting material. These are precisely the activities where AI-assisted retrieval and document analysis may be most useful.
The National Artificial Intelligence Center presented modelling indicating substantial reductions in case-preparation time for some categories of proceedings. It suggested that simpler files could potentially be prepared in minutes and that even much larger cases could see significant reductions in preliminary document work. Those figures remain projections or project-level estimates and should not be interpreted as demonstrated productivity improvements across Azerbaijan’s judicial system.
Using those assumptions, the project team estimated that hundreds of thousands of judicial working hours could potentially be released and described the resulting capacity as equivalent to roughly 167 additional judges. The figure does not represent actual judicial appointments. It is a modelling exercise intended to demonstrate how administrative automation might increase the effective capacity of the existing court system.
That is an important distinction because AI productivity should not be confused with judicial capacity in the institutional sense. Adding a judge increases the number of legally authorised decision-makers. Automating case preparation only increases the amount of information existing judges may be able to process. The two are not identical, even if both can influence the speed at which cases move through the courts.
The project also raises an important question about feedback and accountability. Under the model described at AI4, judges would be able to accept, modify or reject AI-generated drafts. Their corrections could then provide information for improving the system over time. Such a feedback loop could make the platform increasingly aligned with judicial practice, although it also creates governance questions about which decisions should be used for training and how the system avoids reinforcing previous errors or inconsistencies.
Transparency will therefore be critical. If an AI system recommends a legal provision or precedent, judges need to be able to see where that information originated and why it was selected. A legal AI system that merely generates plausible text would be unsuitable for this type of application. Traceability to official legislation and case material is likely to be one of the fundamental requirements for broader judicial deployment.
There are also significant questions about bias and consistency. An AI system trained on historical court decisions can reproduce patterns embedded in those decisions. That does not necessarily mean the underlying precedent is incorrect, but it makes independent judicial scrutiny essential. Automated similarity between cases should support legal analysis rather than become a shortcut that predetermines outcomes.
Azerbaijan’s broader strategy is relevant because the legal projects are not being developed in isolation. The country is simultaneously investing in AI education, research capability, computing infrastructure and national standards. Its 2025–2028 strategy explicitly seeks to develop domestic expertise rather than relying exclusively on technology imported from abroad.
The National Artificial Intelligence Center also operates initiatives including an AI Academy and Datarace.ai, a competition platform through which public and private organisations can present problems for participants to solve. The intention is to create a domestic pipeline connecting education, practical public-sector problems, computing resources and deployable AI applications.
That approach reflects a broader trend likely to become increasingly important for governments. Countries are beginning to treat AI capability not merely as a collection of software products but as national infrastructure involving data centres, computing capacity, skilled workers, standards and trusted public-sector applications.
Justice could become one of the clearest tests of whether that infrastructure can be used responsibly. In sectors such as marketing or customer service, an incorrect AI answer may be inconvenient. In a court system, errors can affect rights, property, liberty and confidence in public institutions. The acceptable threshold for mistakes is therefore much lower.
Azerbaijan’s experiment is significant precisely because it separates different levels of automation. Citizens can use AI to navigate legislation. Judges can potentially use it to accelerate document preparation and legal research. But the authority to interpret the law and make the final decision remains human.
Source: CIJ.World Research & Analysis Team
