CDDH-TE(2026)03

21/09/2026

STEERING COMMITTEE FOR HUMAN RIGHTS

(CDDH)

 

DRAFTING GROUP ON THE USE OF NEW TECHNOLOGIES BY THE COURT [AND NEW TENDENCIES IN LITIGATION BEFORE THE COURT]

(CDDH-TE)

Draft report on the issue of the Court’s use of new technologies

I.              Introduction

1.             The CDDH revised terms of reference for 2026-2027 require it to prepare a report on the use of new technologies by the European Court of Human Rights [and a report on new tendencies in litigation before the European Court of Human Rights] by 31 December 2027. At its 103^rd meeting (24-27 November 2025) the CDDH adopted terms of reference for a Drafting Group (CDDH-TE) to conduct preparatory work for this report [these reports].

2.            At its first meeting (11-13 May 2026), CDDH-TE held an exchange of views [on the use of new technologies by the European Court of Human Rights (Court)] with Veronika FIKFAK (Professor of Human Rights and International Law, Director UCL Institute for Human Rights, Vice President of the European Society of International Law), Laurent PETTITI (Council of Bars and Law Societies of Europe (CCBE)), Klaudiusz RYNGIELEWICZ and Francis DOHERTY (Registry of the European Court of Human Rights), Jan SPOENLE ((Germany), Member of the European Commission for the Efficiency of Justice (CEPEJ) working group on Cyberjustice and AI), and Teresa RODRÍGUEZ DE LAS HERAS BALLELL (President of the European Law Institute, Full Professor (Catedrática) of Commercial Law, Universidad Carlos III de Madrid). [It also held an exchange of views on new tendencies in litigation before the Court with the following experts from the Registry: John DARCY, Klaudiusz RYNGIELEWICZ, Hasan BAKIRCI and Ugur ERDAL.]

II.            The use of new technologies in the justice sector

3.            At a time when many courts are challenged by backlogs, and complex procedural requirements may impede access to justice for individuals, digital technologies have significant potential to improve the accessibility, efficiency and responsiveness of justice systems, and enable them to deliver services better tailored to users’ needs.[1]

4.            While artificial intelligence (AI)[2] is only one form of digital technology, in recent years considerable advancements in generative AI[3] have expanded the range of AI systems[4]  being used in the justice sector.[5] AI systemscan help reduce administrative bottlenecks by analysing incoming applications, categorising them based on complexity and urgency, assigning them to appropriate departments and judges,[6] and even conducting an instantaneous assessment of admissibility criteria.[7] They can also prepare automated transcriptions of hearings,[8] and automated translations, where necessary.[9] AI systems can assist lawyers and judges by analysing documents to create summaries,[10] and support legal research and drafting.[11] AI systems can also facilitate access to courts for individuals through the provision of information systems, such as chatbots, and by reducing legal and linguistic barriers.[12] Additionally, as AI systems can provide insights into court performance and identify inconsistencies in decision-making,[13] they have the potential to improve the predictability of the application of the law and the consistency of court decisions.[14]

5.            The CEPEJ has classified AI systems currently being used in the justice sector using the following categories, in order of frequency of use.[15]

a.    Decision support: AI systems that facilitate or automate stages in the decision-making processes in the justice system, for example, by summarising texts, extracting specific information in applications, and calculating scales for sentencing and compensation.

b.    Document search and review.

c.    Triaging, allocation and workflow automation.

d.    Anonymisation and pseudonymisation.

e.    Information and assistance services: AI systems that provide individuals with information on services available in justice systems and connect them to those services.

f.     Recording, transcription and translation.

g.    Online dispute resolution: AI systems used for the resolution of disputes between parties with limited human intervention.[16]

h.    Prediction of litigation outcomes: AI systems that learn from large datasets to identify patterns in the data that are consequently used to visualise, simulate or predict new litigation outcomes.

6.            The CEPEJ has received no reports of AI currently being used for judicial decision-making, without human supervision.

III.           Risks surrounding the use of new technology in the justice sector

7.            There is evidence that the use of AI systems by national courts has accelerated legal processes, improved judicial responsiveness and broken down barriers that have kept users from engaging fully with the justice system.[17] However, there are also risks associated with the use of AI systems, which the OECD has classified as follows.[18]

a.    Risks flowing from the operation of AI systems (operational risks).

b.    Risks to fundamental rights protections (ethical risks).

c.    Risk that individuals will be excluded from judicial systems by the use of AI (exclusion risks).

d.    Risk that public resistance will undermine trust in judicial systems (public resistance risks).

e.    Risk of missing out on real-world benefits of AI systems by delaying action (risk of inaction).

8.            These risks are addressed in detail in Appendix 1.

9.            The European and International standards relevant to the use of AI in the justice sector are outlined in Appendix 2. However, while the Court may have regard to these texts when deciding on applications before it, it is not, itself, bound by these standards.

IV.          The use of new technology by the Court

10.          The package of reform measures implemented in the decade following the Interlaken Declaration[19] enabled the Court to reduce considerably the number of applications pending before it.[20] While continuation of that reform process has allowed the Court to make further progress in reducing its backlog, as of 31 December 2025 the number of pending cases was 53,450.[21] Efficiency therefore continues to be an essential prerequisite if the Convention system is to remain sustainable,[22] and in its most recent annual report the Court affirmed its dedication to modernising its procedures through digital innovation in order to foster a more efficient, transparent and user-friendly judicial system.[23] However, accountability is also a priority for the Court, which it views as an indispensable corollary to judicial independence, judicial responsibility, and transparency.[24] Consequently, the Court has publicly stated that it is taking a careful and responsible approach to incorporating emerging technologies, centred around tailored solutions to address its specific needs.[25]

11.          The digital strategy implemented by the Court distinguishes between automation and AI.[26] For the Court, “automation” refers to systems that follow pre-defined rules, whereas “true AI” refers to systems that have automatic machine learning capabilities, and are capable of carrying out essentially “human” tasks such as learning, reasoning and decision-making.

1.     “Automation” technologies

12.          The Court uses automation in its Court Management Information System (CMIS) and its Document Management System (DM).

13.          Every new application to the Court is entered into the CMIS database, which records information such as the names and contact details of the parties’ representatives, the applicant’s complaints, “tags” (such as “IMPACT”) and “subjects” (corresponding to the nature of the complaints), the Registry lawyer(s) responsible for the case, and the judicial formation to which it has been assigned. The various procedural events are also recorded in CMIS, and a barcode module allows for the automatic recording of correspondence, case tracking and electronic archiving.  CMIS therefore offers users quick and easy access to all the details of the case and –  as DM is linked to CMIS – all the documents in the case file. Its search function allows users to find cases by reference to defined criteria, and to produce management reports and statistics.

14.          For finished cases, CMIS has a module for the Execution of Judgments, which enables the sharing of information between the Registry and the Secretariat of the Committee of Ministers. The system also manages the agendas of the Committee of Ministers and records resolutions relating to the enforcement of judgments.

15.          DM allows users to link documents to the electronic case files. DM also indexes documents automatically, allowing users to perform full text searches.

16.          The Court has been using templates since 1996 to automate the creation of letters and documents from information entered into CMIS. There are over 2,000 Court-specific templates and macros in 40 languages which are specific to the Court and help Registry staff quickly produce letters and standard documents. A pop-up screen with question prompts helps staff identify the correct template and allows for the automatic completion of certain parts of the text. Once completed, documents are transmitted electronically, using case management “workflows”, and they may be signed using an electronic signature.

17.          The Court communicates electronically with the Governments of the Member States, via their own dedicated secure websites;[27] and with applicants, via eComms.[28]

18.          Since October 2022, the Court has provided a specialised platform for the submission of requests for interim measures under Rule 39 of the Rules of Court,[29] which can be accessed from a computer or mobile device. Previously, these requests had to be submitted by fax or post. In 2024 1,104 requests were submitted through this channel, out of a total of 1,630 received.

19.          The Court disseminates information about its case-law and communicated cases via the Hudoc database.[30] The Court's Jurisconsult Department has also developed the Court's Knowledge Sharing Platform (ECHR-KS)[31] in collaboration with the Directorate General of Human Rights and Rule of Law. The Platform is updated weekly and provides direct access to developments in the Court’s case-law presented thematically, by reference to Articles of the Convention and by transversal themes. In addition to the documents prepared by the Registry, there are documents and links of more general jurisprudential interest. The ECHR-KS is available in both official languages, English and French, as well as in three non-official languages: Romanian, Turkish and Ukrainian.

2.    AI systems

20.          The Court is currently employing AI systems in a number of different areas and is testing/developing others. It is also working on two major projects into which AI systems will in future be integrated.

21.          AI systems being used and developed by the Court will support and improve upon existing automation systems. They are being explored strictly for administrative and organisational purposes, and not with a view to their having any role in the Court’s decision-making processes.[32]

1.1.  AI systems currently in use

a.    Cybersecurity

22.          The Court uses AI to strengthen its cybersecurity. An AI system (“SentinelOne”) scans servers and laptops for unusual activity, while another system (“Security Operations Center”) can detect, analyse and respond to cyberthreats in real time.

b.    Translation and language systems

23.          The Court uses AI-powered systems such as “SYSTRAN”, “Antidote Corrector” and “DeepL” to boost efficiency in translation and language correction. SYSTRAN is an AI-based Neural Machine Translation (NMT) system used by professional translators in the Court’s language department, and according to a two-year proof of concept study it has boosted efficiency by approximately 10 percent. Antidote Corrector and DeepL are also used by translators and/or language checkers to assist with linguistic quality. These systems assist – but do not replace – translators and language checkers. Given the potential consequences of inaccurate translations of legal texts, the Court considers the intervention of a qualified professional to be essential.

c.    AI chatbot

24.          The Court uses Microsoft Copilot, a general-purpose AI chatbot that draws on publicly available information rather than Court data. The Court has an enterprise agreement with Microsoft, which offers Registry staff enhanced data protections when using this chatbot with their work account. As a consequence of these enhanced data protections, chat data is encrypted, meaning that Microsoft has no access to users’ data. Chat prompts and responses are not used for AI training purposes, nor are they available for human review.

d.    Video production

25.          AI software such as Descript is used by the Court to produce video tutorials. The software transcribes speech into text and enables videos to be edited by editing the transcribed text.

e.    Coding

26.          The Court’s IT Department uses AI agents in its development team. It has both GitHub  Copilot and Claude code licences and is using agentic AI[33] for software development. AI agents generate code snippets, which are then reviewed by (human) developers.

f.     Speech-to-text

27.          Registry staff have access to the Council of Europe’s “speech-to-text” application, which converts speech (for example, in video or audio recordings) to text.

1.2.  AI systems currently being tested or developed (“Labs projects”)

28.          AI systems in this category are being researched within the Court. While they are still at an experimental stage, many are already in use internally.

a.    Court AI Assistants

29.          The Court AI Assistants are custom-developed, private chatbots which can answer questions on a specific area.[34] There are currently three examples: the Case Processing Manual Assistant; the IT Assistant; and the Address Verification Assistant.

30.          The Case Processing Manual Assistant was launched in November 2025 and is currently being tested within the Registry of the Court. The Assistant has been fed approximately 400 of the Court’s case-processing manuals.[35] Registry lawyers and administrative assistants may ask it procedural questions (for example, “how do I launch a Rule 39 checklist?”), and follow-up with further questions based on the Assistant’s response (for example, “when should I clone a Rule 39 checklist instead of creating a new checklist?”). Approximately one third of administrative Court staff have used the Case Processing Manual, and there are approximately 25 daily active users.

31.          The IT Assistant, which will pull knowledge from helpdesk and IT manuals, and the Address Verification Assistant, which will verify the accuracy of postal addresses from a wide range of countries, are still in development.

b.    Hudoc

32.          Hudoc is currently powered by Elasticsearch, which relies on keyword-based searching. Elasticsearch now supports semantic search, which weighs the meaning behind a query rather than its exact wording and returns the most relevant results. The Court does not currently use semantic search on Hudoc, but a proof of concept is underway.

33.          In addition, an experimental project is underway that will allow key concepts from documents published on Hudoc to be extracted by AI, and users will be able to filter/carry out searches using these concepts. Users can currently filter/carry out searches using keywords assigned to cases by Registry lawyers, but these keywords are mostly technical legal terms. The key concepts are intended to be more user-friendly for non-lawyers.

 

c.    Video conferencing

34.          AI can assist in summarising meetings, generating action items, transcribing conversations and translating in real time. The Court’s IT department is currently trying out a new feature which will allow users to give a voice sample in order to improve the quality and accuracy of AI’s ability to identify who is speaking in a meeting.

d.    Infographics

35.          AI can identify key information in documents and present them visually. This technology will be used to transform standard manuals into an infographic style, which will show only information specific to the user.

e.    Other projects

36.          These include text-to-speech software, transforming written text into fluent, natural sounding speech; Optical Character Recognition, converting complex handwritten or printed documents across multiple languages into editable and searchable text; and improving the efficiency of IT Helpdesk by allowing AI to manage tickets and suggest replies.

1.3.  Major projects which will integrate AI systems

37.          There are two major projects in development at the Court which, while not centred around AI, will have AI systems integrated into them at a later stage of development.

a.    The case processing Gateway

38.          Once completed, the “Gateway” will be a platform dedicated entirely to case management. Whereas CMIS previously had separate systems for each stage in the life cycle of an application, and information had to be inputted afresh at each stage, the new Gateway will improve upon the existing CMIS database in order to streamline the case management process. It will collect all the data necessary to qualify the legal issues raised by an incoming application and direct it either to a Single Judge, a Committee or a Chamber of seven judges. Once data is collected, the application may be dealt with without delay. Even after a case has been notified (communicated) to the respondent Government, the case processing Gateway will guide the lawyer through the process of drafting the decision or judgment. It will provide the lawyer with all the available information about the state of the Court’s case-law, information on the execution of the Court’s judgments, and, eventually, all of the information contained in national databases (such as case-law and legislation databases). Throughout the essential stages of the procedure, the information contained in the Jurisconsult’s Knowledge Sharing Platform will also be made available to the lawyer to ensure that the case is dealt with in the most coherent, consistent and efficient way.

39.          Automation will drive this process, and it is envisaged that in future the AI systems currently being used and developed by the Court will be integrated into the Gateway.

40.          The procedures for managing manifestly inadmissible and repetitive requests have already been integrated into the Gateway (being, respectively, the first and second completed phases of the project).

41.          In respect of manifestly inadmissible requests, there are three components to the system developed in the first phase of the Gateway project: the “Single Judge module” in the CMIS Gateway portal; a workflow; and documents (the Single Judge decision and cover letter). The case processing lawyer examines the incoming application form, complaint by complaint, and enters standardised information into the CMIS Single Judge module concerning the complaints and proposed inadmissibility grounds. This standardised information is accompanied by a more detailed (and case specific) analysis of the facts of the case and the admissibility of the complaints. Once the relevant information has been entered into the Single Judge module, the lawyer can send the proposal to the Single Judge for validation via an electronic workflow. When the decision has been taken by the Single Judge, the Single Judge decision and cover letters are generated. The decision is generated in English or French, while the cover letters are generated in the applicant’s own language.

42.          For repetitive cases, the well-established case-law fast track module (WECL FT module) has been developed to allow the Court to manage, on a “one in one out” basis, repetitive cases linked to structural problems, for which there is well-established jurisprudence. The WECL FT module is used to record information concerning the subject matter of the application and the objections to be brought to the attention of the respondent Government, as well as the complaints to be declared inadmissible at the stage when the complaint which is the subject of well-established case-law is notified (communicated) to the respondent Government. It records all relevant subsequent information, and, subject to the oversight of Registry lawyers, it calculates the amounts of just satisfaction, and generates automatically draft letters, friendly settlement declarations, reports, decisions and judgments. The WECL FT module can be adapted to the specific issues arising in each respondent State. There are currently 205 WECL FT modules that cover all Convention Articles. 

43.          The Court is currently in phase three of the Gateway project, being the development of the interface necessary for the notification of well-founded applications to respondent Governments.

b.    The online application form

44.          The online application form project aims to develop a secure, user-friendly, multilingual interface allowing members of the public to complete and submit online applications to the Court.

45.          A prototype of the online application form will shortly be tested internally, before Bar associations and NGOs are invited to take part in an external test. The first prototype is an exact copy of the paper application form and has no AI features. However, AI research is an integral part of the project and AI features will be added in future, for example, for the purposes of translating or summarising incoming applications.

46.          When the online application form is operational, its use will not be compulsory. Applicants and their representatives will have a choice between submitting an application form online or submitting it by post. Users of the online application form will be authenticated at the point of submission, and there will be controls on the volume of applications that may be submitted from a single account.

3.    Safeguards

47.          The Court is subjecting its use of AI systems to four fundamental guidelines:

a.    Human oversight is essential: AI must assist, but not replace, judicial decision-making.

b.    Transparency and accountability: AI use should be communicated openly, ensuring full accountability for the results.

c.    Strong data protection: safeguarding sensitive legal information is a priority.

d.    Clear policies on AI use and continuous review: AI must be regularly assessed for performance and risks to uphold standards. 

48.          In furtherance of these guidelines, the following safeguards are employed by the Court.

1.1.  Security of data

49.          Data security and privacy protections are key considerations for all the Court’s AI systems. The AI Assistants have been custom developed (see paragraphs 29-31 above). The Microsoft Copilot chatbot is subject to an enterprise agreement with Microsoft, which offers Registry staff enhanced data protections when using it (see paragraph 24 above). SYSTRAN (see paragraph 23 above) is a cloud translation service run by a French provider on French data centres. The connection between the Court and SYSTRAN is encrypted and the system is trained solely on public HUDOC material.

1.2.  Human oversight

50.          The Court views AI as an “intelligent assistant” and not as a means of replacing judicial decision-making – or, indeed, as a means of replacing any members of staff. It has a clear policy that when AI systems are used, there must always be a human in the loop. While in future AI systems might be used to assist Registry lawyers with technical matters, it is envisaged that they will do so by drawing certain issues to the lawyers’ attention (for example, possible non-compliance with the Rule 47 criteria or the four-month time-limit). AI systems will not be employed to take decisions, even on technical matters, and the Court has no plans to use them to assist lawyers and judges with substantive legal decisions (for example, proportionality).

1.3.  Transparency and accountability

51.          The Court has an intranet site (CourtAI) dedicated to providing Registry Staff with information about its use of AI, and guidance on the use of AI systems. CourtAI sets out the four fundamental AI guidelines which govern the Court’s use of AI systems (see paragraph 47 above). In addition, it sets out four key principles to guide Registry staff using any AI sites and services at the Court. Those four key principles are as follows:

a.     Security, privacy and data protection: Court documents or data that are not already publicly available must not be used with Copilot, ChatGPT or other, similar AI services.

b.    Always fact-check: be aware that responses from AIs may be inaccurate or biased. Never copy and paste AI-generated output into official documents. At a minimum, double-check any results against another source and carefully review.

c.     Use discretion: AI is a tool. It is not always the right tool. Consider carefully whether it is right for the purpose to which you are planning to apply it.

d.    Accountability: you are ultimately responsible for the quality and accuracy of everything you do.

52.          Information about the Court’s use of AI is also publicly available. In addition to this document, details may be found in the Court’s Annual Report,[36] and in the published proceedings of a Conference in June 2025 on the system of the European Convention on Human Rights.[37]

53.          The use of AI by the Court is overseen at three different levels: there is a Council of Europe AI Working Group; a Court AI Working Group; and a Judges’ IT Working Party. These working groups are closely involved in all AI-related projects to ensure that they are and remain in line with the values and operational needs of the Court.

54.          If the use of new technology by the Court would require a change in the Rules of Court, the Contracting States must first be invited to submit written comments on the proposals, and the Registrar must also invite written comments from organisations with experience in representing applicants before the Court and from relevant Bar associations.[38]

1.4.  Training and resources

55.          The Court runs regular trainings on AI systems. For example, there have been multiple training sessions for Registry staff on the new Case Processing Manual Assistant, and they will shortly be offered training on the use of the Microsoft Copilot chatbot.

56.          In addition, the IT Department is working on a “prompt library”; that is, a library of reusable prompts for users.

1.5.  Exchanges with other international courts and treaty bodies

57.          The Court holds regular exchanges with the Inter-American Commission on Human Rights, the African Court of Human Rights, the Court of Justice of the European Union and treaty bodies in Geneva to discuss, inter alia, the use of new technology.

4. Impact

58.          While some benefits can already be seen, the full impact of using new technologies has not yet been felt. Nevertheless, the changes across a number of areas are likely to be profound.

1.1.  Efficiency

59.          The use of automation has already allowed the Court to enhance the efficiency and effectiveness of its internal and external processes. The use oftemplates has allowed the Court to manage the considerable flow of correspondence that it receives every year. In addition, the implementation of the new Single Judge module has helped the Court clear the backlog of inadmissible cases pending before 2011, and also accelerated the processing of such applications. Today, an applicant can count on receiving an inadmissibility decision within an average of three to six months from the registration of the application. The introduction of the WECL FT module has allowed the Court to concentrate on the treatment of priority cases and those which raise important or even new issues from the perspective of the Convention. It has also shortened processing times to one year from the time a (repetitive) case is notified to the Government to a final ruling, at least in part because observations are not required from the respondent Governments. The launch of the online application form (see paragraphs 44-46 above) and completion of the additional phases of the Gateway project (see paragraphs 38-43 above) are likely to further improve the efficiency of the Court’s case-management processes, especially when AI systems are integrated so as to support and enhance existing automation systems.

60.          There have also been efficiency gains in areas in which AI systems are currently being employed by the Court, most notably in the fields of translation (see paragraph 23 above) and software development (see paragraph 26 above).

61.          In addition, the deployment of AI systems in the field of cybersecurity has improved the Court’s ability to detect and respond to cyberthreats (see paragraph 22 above).

1.2.  Budget

62.          In order properly to harness AI and ensure its integration is a success, continuous investment is required, both in IT and in training at the human resource level. This poses certain challenges, especially as the prices of equipment required for AI (for example, servers, RAM,[39] Central Processing Units and Tensor Processing Units[40]) have increased rapidly, which will impact the Court’s IT budget for at least the next 18 months, or until supply catches up with demand.

1.3.  Staffing

63.          As AI systems increasingly take over human tasks, there will be inevitable consequences at a human resources level. This is already being felt in the IT Department of the Court, where the use of AI agents for software development has fundamentally changed the work of (human) developers (see paragraph 26 above) and consequently impacted upon hiring and training decisions. In addition, automation and AI systems have taken over some of the tasks of approximately 150 administrative assistants working at the Court and will likely encroach further in the coming years. To date, the Court has responded to this challenge by retraining administrative assistants and finding new tasks for them. One example of this is the “paralegal project”, which involves training administrative assistants to assist Registry lawyers by carrying out preliminary Rule 47 checks and identifying repetitive complaints in incoming applications.

64.          The Court is working together with the wider Council of Europe in order to discuss the need for “reskilling” and “upskilling” staff whose tasks have been taken over by AI systems.

V.            The use of new technology by parties to litigation before the Court

65.          While the focus of this report is on the use of new technology by the Court, applicants are also using new technology – and AI in particular – to assist them in lodging applications with the Court. While the responsible use of AI by applicants and their representatives would not necessarily be a cause for concern, there are a number of problems that might arise.

66.          First of all, the risk of AI hallucinations will increase the risk of inaccurate information being submitted to the Court, even by applicants acting in good faith. While this problem might arise both in applications submitted by lawyers and applications submitted by litigants in person, the latter are more likely to use AI systems available on the wider market, and are less likely to detect hallucinations. Applicants might also act in bad faith, for example, by seeking to manipulate AI systems through hidden prompts or instructions embedded within documents,[41] or by using AI to fabricate evidence in support of their applications.[42]

67.          The Court’s Guidance Notes for Filling in an Application Form[43] now includes the following text:

“Use of artificial intelligence tools when filling out the application form

If you make use of artificial intelligence (AI) tools to help you prepare your application form, you remain fully responsible for the content submitted to the Court. You are therefore advised to check that all factual information, legal arguments and references to case-law or other sources generated or suggested by such tools are accurate, relevant and complete.

Please note that knowingly submitting untrue facts, false declarations, or incomplete and therefore misleading information may result in the application being declared inadmissible as an abuse of the right of application under Article 35 § 3 of the Convention."

68.          Secondly, there is a risk that the use of AI will make it easier for applicants to generate a large number of applications, thereby flooding the Court. This risk will be particularly acute when the online application form is available to litigants in person. While the Court has its fair share of abusive applicants, who submit repeat, unmeritorious applications, the need to (i) complete an application form and (ii) post a hard copy of that form – and any supporting documents – to Strasbourg currently provides a check on such behaviour. When an application can be completed using AI, and uploaded to an online platform, the number of applications received from abusive applicants is likely to increase significantly. It is for this reason that the online application form has been designed with technical counter-measures intended to limit mass and automated filing (see paragraph 46 above), and its use will be monitored to identify coordinated or automated campaigns. However, while such measures may mitigate the risk, they will not eliminate it. As all incoming applications have to be reviewed by a Registry lawyer, including those submitted by abusive applicants, any significant increase in the volume of applications will have resource implications for the Court.

69.          Not every abusive applicant acts in bad faith, with the express intention of derailing the work of the Court. However, the inadmissibility decision in Zambrano v France[44]provides a timely warning of the difficulties that an abusive applicant can pose to the Court, even without AI or the ability to submit an application form online. The applicant in that case sought to challenge France’s use of vaccines and the creation of a health-pass in response to the Covid-19 pandemic. He encouraged visitors to his internet site to send a copy of his application to the Court in order to paralyse it and derail the whole system. As part of the applicant’s initiative, almost 18,000 applications were lodged with the Court.

VI.          Conclusion

70.          While the numerous reforms introduced during and after the Interlaken Process have reduced significantly the backlog of cases pending before the Court, new technologies, including AI, now offer it an opportunity to achieve even greater efficiency gains. This is a particularly important consideration, given that the roll-out of the online application form, coupled with the use of AI by applicants, has the potential to considerably increase the number of applications the Court receives. New technologies could also offer considerable benefits to applicants and their representatives, through the provision of information tools, the reduction of legal and linguistic barriers, more efficient case-processing, and simplified channels of communication with the Court.

71.          There are undoubtedly considerable risks associated with the use of AI systems within the justice sector. Of particular concern is the need to ensure the fundamental rights of applicants, including their right to a fair trial and the protection of their data. There are also operational risks, including the risk of inaccuracies, hallucinations and bias, and increased vulnerability to cyber-attacks. In view of the high stakes involved in judicial decision-making,[45] those risks must be identified and mitigated in order to ensure that justice is fair and transparent, that fundamental rights are respected, and that confidence in the justice system is not eroded.[46]Nevertheless, a failure by the Court to engage with new technology would carry the dual risk of losing out on its real-world benefits, while encouraging Registry staff and judges to engage in the use of shadow AI, which carries all the aforementioned risks without any safeguards.

72.          Having regard to the foregoing considerations, the CDDH notes that the Court is both dedicated to modernising its procedures through digital innovation, and at the same time committed to taking a careful and responsible approach to incorporating emerging technologies, centred around tailored solutions to address its specific needs and the needs of its users.

73.          In the specific context of the use and development of AI systems, the CDDH recognises that the fundamental principes governing the use of these systems by the Court (see paragraphs 47 and 51 above) are consistent with European and International standards on the use of new technology in the justice sector (see Appendix 2). In particular, the CDDH recognises that:

a.    data security and privacy protections are key considerations for the Court;

b.    the Court has stated that AI may assist, but will not replace, human decision-making;

c.    the Court is using and developing AI systems strictly for administrative and organisational purposes, and not for judicial decision-making;

d.    while AI systems may, in future, assist judges and lawyers with technical (as opposed to substantive) legal matters, such assistance will always be subject to human oversight; and

e.    the Court is committed to being transparent about its use of AI systems.

74.          Finally, the CDDH notes that the Court is planning for the future use of new technologies, and in particular AI, by investing in IT equipment and in training. It is also working with the Council of Europe to discuss the need for “reskilling” and “upskilling” staff whose tasks have been taken over by AI systems. The CDDH recognises, however, that if the Court is to properly harness AI and ensure its integration is a success, continuous investment will be required.

APPENDIX 1

Risks surrounding the use of new technology in the justice sector

1.    Operational risks

1.1.  Accuracy and explainability

1.             As generative AI systems learn patterns and characteristics from large collections of data,[47] they are highly sensitive to the quality of the training data and may amplify errors embedded in the datasets.[48] AI systems are also susceptible to overfitting patterns and learning spurious correlations.[49] Hallucinations occur when AI systems invent information and present it as fact.[50] While hallucinations may be mitigated, they cannot be eliminated entirely.[51]

2.            AI decisions may be difficult to understand (sometimes referred to as a lack of “explainability”). Even AI experts, including the developers of AI systems, may not understand how their outputs are produced. This opacity, or “black box” effect, may hide deficiencies in AI systems, such as the existence of bias, inaccuracies, or hallucinations.[52]

1.2.  Security

3.            AI systems are vulnerable to cyber threats, which may lead to data breaches, privacy violations and loss of functionality.[53] It is also possible for malicious cyber actors to manipulate AI systems, altering their outputs and causing erroneous decisions or actions.[54]

1.3.  Suitability for specialist domains

4.            Legal language constitutes a technical domain with specialised terminology that may diverge significantly from the meaning in common usage.[55] Commercial AI systems may struggle to use this terminology accurately, interpret context correctly, and understand the nuanced reasoning patterns characteristic of judicial work.[56] In addition, the law, and in particular human rights law, is constantly evolving.[57] As generative AI systems generate new content that mimics the characteristics of the original training data,[58] they are ill-suited to a dynamic and evolutive interpretation of the law.[59]

1.4.  Human frailty

5.            Humans are susceptible to automation or confirmation bias, meaning that they tend to place unjustified trust in machines and select information that supports their own beliefs. They may therefore validate the output of AI systems without sufficiently questioning it, especially when it corresponds with their own views.[60] Such overdependence on AI by lawyers and judges could lead to a decline in critical thinking and drafting skills.[61] At the other end of the spectrum, humans may disproportionately mistrust AI systems, leading them to discount their recommendations when they should be followed.[62]

2.    Ethical risks

6.            Insofar as the ethical risks outlined below present potential challenges to fundamental rights protections, including rights guaranteed under the Convention, these challenges are, for the time being, somewhat speculative as the Court has not yet had the occasion to consider any Convention issues arising from the use of an AI system.

1.1.  Fair trial guarantees

7.            The integration of AI systems into judicial processes could challenge the fair trial guarantees in Article 6 of the European Convention on Human Rights (Convention).[63]

8.            On a practical level, changes to the means of lodging an application and/or communicating with a court could impede an individual’s right of access to court. The Court has held that a requirement that applications be made electronically, while disregarding the hurdles faced by a particular individual in complying with that requirement, was overly formalistic and amounted to an unjustified interference with the right of access to court.[64]

9.            AI systems responsible for the initial categorisation of a case may determine the pathway it continues along.[65] This could impact on a litigant’s right to adversarial proceedings,[66] and perhaps even predetermine the outcome of the case.[67] The principle of equality of arms might also be prejudiced if the parties do not have equal access to information about the use of AI systems.[68]

10.          Given the risk of inaccuracies and hallucinations, the use of AI systems for research or decision support could pose risks to legal certainty and the rule of law.[69] Furthermore, opaque decision-making processes and non-explainable recommendations resulting from the use of AI systems may not satisfy the requirement that a court provide adequate reasons for its decisions.[70]

11.          The use of AI systems that predict case outcomes could indirectly pressure judges to conform to statistical models, undermining judicial autonomy.[71]

12.          More globally, the transfer to AI systems of responsibility for decisions that might impact an individual’s fundamental rights raises important ethical questions since human decision-makers – in contrast to AI systems –  are endowed with empathy.[72] While human oversight (the human in the loop) provides an important safeguard, the effectiveness of this safeguard might be diminished where deficiencies in AI systems are masked by a lack of “explainability” (see paragraph 2 above)[73] or where the human in the loop suffers from automation or confirmation bias (see paragraph 5 above).[74]

13.          While any unfairness before a lower court could be corrected on appeal, any unfairness before a court of last instance could deny individuals an appropriate remedy for their complaints. This is of even greater concern for international courts and tribunals; for example, decisions and judgments of the Court – which provides individuals with a final opportunity to hold States accountable for violating their Convention rights – are not subject to further review.[75]

1.2.  Bias

14.          AI systems do not operate autonomously; they are shaped by human choices at every stage, from model selection and training data to fine-tuning and parameter adjustments.[76] Furthermore, the training data itself is never neutral, as it embeds all biases, inaccuracies, gaps or failures contained in the training database and/or the cultural biases of those who designed the system and guided its training.[77] AI systems may therefore be built upon and sustained by data and models that reproduce, perpetuate and exacerbate existing bias and prejudice.[78] Consequently, the use of these systems in the justice sector could raise issues under Article 14 of the Convention (which provides that the enjoyment of the rights and freedoms set forth in the Convention shall be secured without discrimination) and Article 1 of Protocol No. 12 to the Convention (which contains a general prohibition of discrimination in the enjoyment of any right set forth by law).[79]

1.3.  Accountability

15.          Given that the use of AI systems entails the delegation of human tasks to machines, it risks obscuring responsibility for the performance of those tasks, thereby undermining traditional mechanisms for holding individuals responsible for the harm they cause.[80] Concerns about accountability are compounded when there is a lack of transparency about the use of AI systems. Unless relevant stakeholders understand the underlying processes and algorithms, it may be difficult to trace and assign responsibility in the event of adverse impacts on human rights, democracy or the rule of law.[81] Accountability gaps resulting from the use of AI systems in the justice sector could therefore hinder the exercise of the right to an effective remedy in Article 13 of the Convention.[82]

1.4.  Privacy and data protection

16.          Data inputted into a commercial generative AI model are usually transmitted to the company providing the service, recorded on their servers, and may become available for model retraining purposes.[83] As courts and authorities involved in the administration of justice handle and retain personal data, including sensitive data,[84] the use of commercial AI systems by these bodies, without the mediation of formal contractual arrangements with the service provider governing the transfer, retention and use of data, might result in the unintended disclosure of classified or otherwise sensitive information,[85] which could interfere with the parties’ rights under Article 8 of the Convention.[86]

3.    Exclusion risks

17.          The digital divide refers to the gap between those who can access and use information and communication technologies and those who cannot.[87] This could be due to a lack of access to the internet, or a lack of digital literacy. The use of AI systems in the justice sectorcv  could exacerbate this digital divide and create inequality in access to justice.[88] 

4.    Public resistance risks

18.          If AI systems are integrated into the justice sector, a lack of public understanding, or misconceptions, about how they are being used, and their capacities, could undermine trust in the justice system and the legitimacy of the courts.[89]

5.    Risk of inaction

19.          Delays in leveraging AI may result in a loss of real-world positive benefits, both in terms of efficiencies and costs.[90] According to one of the most senior judges in the United Kingdom, “lawyers and judges have no choice but wholeheartedly to embrace and adopt the use of artificial intelligence so as to make their work outputs more efficient, cost-effective and more quickly deliverable”.[91]

20.           Failure to leverage AI could also increase the risk of unauthorised use of AI by employees (shadow AI).[92] In contrast to the authorised use of AI, shadow AI is not subject to governance and carries a heightened risk of data protection and security breaches.  

APPENDIX 2

European and International standards relevant to the use of new technology in the justice sector

1.    International treaties

1.1.  AI

1.            The Council of Europe’s Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law (the Framework Convention), which was opened for signature on 5 September 2024, is the first international treaty on AI and human rights.[93] Each party is required to adopt or maintain measures related to activities within the lifecycle of AI systems that ensure compliance with the protection of human rights;[94] the integrity, independence and effectiveness of democratic institutions and processes, including the principle of the separation of powers, respect for judicial independence and access to justice;[95] respect for human dignity and individual autonomy;[96] transparency and oversight;[97] accountability and responsibility for adverse impacts;[98] respect for equality and the prohibition of discrimination;[99] and privacy rights and data protection.[100] Each party is also required to ensure the availability of accessible and effective remedies for violations of human rights resulting from activities within the lifecycle of AI systems.[101] To this end, they must ensure that relevant information regarding AI systems which have the potential to significantly affect human rights is documented, provided to bodies authorised to access that information and, where appropriate and applicable, made available or communicated to affected persons;[102] that this information is sufficient to contest the use of an AI system or the decisions made or informed by it;[103] and that there is an effective possibility for persons concerned to lodge a complaint to the competent authorities.[104]

2.            The EU AI Act[105] entered into force on 1 August 2024, and became fully applicable on 2 August 2026, with some exceptions. The AI Act categorises AI systems based on potential risk level (unacceptable, high, limited and minimal) and fixes obligations accordingly. AI systems intended to be used by a judicial authority or on its behalf to assist the judicial authority in researching and interpreting facts and the law and in applying the law to a concrete set of facts, or to be used in a similar way in alternative dispute resolution, are deemed to be “high-risk AI systems”.[106] According to recital 61, while the use of AI tools can support the decision-making power of judges or judicial independence, it should not replace it. The final decision-making must remain a human-driven activity. However, recital 61 further clarifies that the “high risk” classification does not extend to AI systems intended for purely ancillary administrative activities that do not affect the actual administration of justice in individual cases, such as anonymisation or pseudonymisation of judicial decisions, documents or data, communication between personnel, or administrative tasks.

3.            The requirements for high-risk AI systems include a risk management system,[107] the training, validation and testing of data sets,[108] technical documentation,[109] record keeping,[110] transparency,[111] human oversight,[112] and accuracy, robustness and cybersecurity.[113] The providers of high-risk AI systems must put in place a quality management system,[114] keep certain documentation at the disposal of the national competent authorities,[115] and cooperate with the competent authorities.[116] Prior to deploying a high-risk AI system, deployers are required to carry out a fundamental rights impact assessment.[117]

1.2.  Human Rights

4.            As the Framework Convention complements existing international standards on human rights, democracy and the rule of law, the European Convention on Human Rights[118] continues to apply. Of principal relevance to the use of AI systems in the justice sector are the right to liberty and security (Article 5), the right to a fair trial (Article 6), the right to respect for private and family life, home and correspondence (Article 8), the right to an effective remedy (Article 13), and the prohibition of discrimination (Article 14 and Protocol 12).

5.            The Charter of Fundamental Rights of the European Union,[119] which applies to all EU institutions and member states when they are implementing EU law, also contains a right to liberty and security (Article 6), a right to respect for private and family life (Article 7), the principle of non-discrimination (Article 20) and a right to an effective remedy and fair trial (Article 47). It also contains a specific right to protection of personal data (Article 8), a right to equality before the law (Article 21) and enshrines the principle of the presumption of innocence and right of defence (Article 48).

6.            The International Covenant on Civil and Political Rights[120] contains a right to liberty and security (Article 9), fair trial guarantees and a principle of equality before the law (Articles 14, 15 and 26), a right to respect for family life, private life, home and correspondence (Article 17), and a principle of non-discrimination (Article 2).

1.3.  Data Protection

7.            The Council of Europe Convention for the Protection of Individuals With Regard to the Automatic Processing of Personal Data[121] provides guarantees in relation to the collection and processing of personal data; outlaws the processing of "sensitive" data on a person's race, politics, health, religion, sexual life, criminal record, etc., in the absence of proper legal safeguards; enshrines the individual's right to know that information is stored on him or her and, if necessary, to have it corrected; and imposes some restrictions on transborder flows of personal data to States where legal regulation does not provide equivalent protection.

8.            The General Data Protection Regulation (GDPR)[122] lays down rules relating to the protection of natural persons with regard to the processing of personal data and rules relating to the free movement of personal data. It protects individuals when their data is being processed by the private sector and most of the public sector, and allows them to better control their personal data. Further procedural rules for cross-border enforcement laid down in Regulation (EU) 2025/2518[123] will apply from 2 April 2027.

2.    Standards and guidelines

9.            The CEPEJ European Ethical Charter on the Use of Artificial Intelligence in Judicial Systems and their environment (Ethical Charter)[124] sets out five fundamental principles.

a.     Respect for fundamental rights, including the right of access to a judge and the right to a fair trial.

b.    Non-discrimination.

c.    Quality and security of data.

d.    Transparency, impartiality and fairness.

e.    User autonomy (“under user control”).

10.          The CEPEJ Guidelines on the use of generative AI by Courts[125] provide a framework for the safe implementation of generative AI in the administration of justice, respecting the rights guaranteed by the Convention, the Framework Convention, and  Convention 108,[126] as well as the principles set out in the Ethical Charter, and, where applicable, the AI Act. They are based on four core principles.

a.    The exercise of judicial power is the exclusive responsibility of the courts.

b.    Effective access to a human judge is always guaranteed.

c.    Outputs produced by generative AI are never binding.

d.    The use of generative AI in judicial activities and the drafting of decisions must be transparent.

11.          The Guidelines are split into operational and normative aspects. Operationally, they recommend that courts use AI only in areas where it is suitable, and in response to an identified problem. They also recommend that courts prefer customised, secure systems over generic, off the shelf solutions; and, when resorting to commercial solutions, the deployment of generative AI should be exclusively mediated through formal contractual arrangements with AI service providers addressing the transfer, use, protection and destruction of data. Finally, they recommend specialist training for AI users and encourage courts to be aware of how the parties might be using AI.

12.          The normative aspects focus on respect for fundamental rights, notably Articles 5, 6, 8, 13 and 14 of the Convention, and Protocol No. 12. They include requirements for judicial independence and mandatory human responsibility and supervision. Additional safeguards include explainability and transparent reasoning, and State liability for damages caused by the use of generative AI.

13.          CEPEJ has also prepared Guidelines on electronic court filing (e-filing) and digitalisation of courts[127] based on fundamental legal principles such as the protection of the legal rights of all users of digital justice, fair trial guarantees, including the independence of the judiciary, non-discrimination, access to an effective judicial remedy and the relevant data protection principles. The principle of “digital by default” is identified as a key element of any e-governance strategy. This means that digital services should be so straightforward and convenient that all who can use them will choose to do so, while those who cannot or do not wish to are not excluded. To this end, the needs of all end-users should be taken into account when designing, delivering, implementing and evaluating digital solutions and services to ensure that they are inclusive, accessible and serve the needs of users.

14.          HUDERIA[128] provides public and private actors with structured guidance for assessing the risks and impacts of AI systems with regard to the protection and promotion of human rights, democracy and the rule of law. It consists of two complementary elements: the HUDERIA Methodology, which sets out a structured approach to risk and impact assessment, and the HUDERIA Model: Context-Based Risk Analysis (COBRA), which supports the systematic collection and mapping of information about an AI system’s context, design and deployment in order to identify potential risks and determine appropriate governance and mitigation measures.

15.          The OECD AI principles,[129] which were adopted in 2019 and updated in 2024, seek to guide AI actors in their efforts to develop trustworthy AI and provide policymakers with recommendations for effective AI policies. They are divided into value-based principles and recommendations for policy makers. The value-based principles focus on respect for the rule of law, human rights and democratic values, including fairness and privacy (Principle 1.2); transparency and explainability (Principle 1.3); robustness, security and safety (Principle 1.4); and accountability (Principle 1.5). Recommendations for policy makers include investing in AI research and development (Principle 2.1) and building human capacity and preparing for labour market transformation (Principle 2.4).

16.          Committee of Ministers Recommendation CM/Rec(2020) 1 to Member States on the human rights impact of algorithmic systems sets out guidelines in an appendix.[130] The guidelines require Member States to refrain from violating human rights through the use of algorithmic systems, and to develop legislative and regulatory frameworks that foster an environment where all actors respect and promote human rights and seek to prevent possible infringements.[131] The human rights impact of all algorithmic systems should be assessed regularly, from proposal and throughout their life cycle. The Recommendation also stresses the importance of transparency, the identifiability of algorithmic decision-making, oversight and access to effective remedies. More specifically, it indicates that the use of algorithmic systems in the judicial field for the purpose of legal analysis, prediction or individual risk assessment should be introduced with great care and in conformity with the guarantees of a fair trial as enshrined in Article 6 of the Convention.