AI in Courts: Global Lessons for Mauritius · Newsletter Article · June 2026
EDITORIALAn editorial on the global acceleration of AI in justice systems — and what Mauritius can learn from the United States, Europe, and Singapore.
Artificial intelligence is no longer a distant prospect. It is reshaping sectors across the world at a pace that demands attention from every institution — including the judiciary. Mauritius is not immune. With the launch of the National Artificial Intelligence Strategy (2025–2029) on 9 April 2026, supported by the United Nations Development Programme (UNDP), and the publication of the FAIR Guidelines — grounded in the principles of Fairness, Accountability, Inclusiveness, Integrity, and Responsibility — Mauritius has signalled its ambition to become a high-tech, innovation-driven "Intelligent Island." [1]
The legal sector is not exempt from this transformation. The question is not whether AI will touch the judiciary — it already has, in ways both visible and invisible. The question is whether the judiciary will shape its adoption deliberately, or allow it to happen by default.
AI systems — more precisely, machine learning models trained on historical legal data — can analyse past decisions, extract patterns, and generate statistical probabilities of how a court may decide on a given matter. This is sometimes called predictive justice: the use of jurisprudential data to forecast likely outcomes. [2]
For practitioners, this can offer a forecast of the likely success rate of a case where a similar dispute has already been adjudicated. If used responsibly, this may encourage earlier out-of-court settlements, reduce protracted litigation, and relieve pressure on court registries. For court administration, AI tools can categorise cases lodged, allocate resources more efficiently, and assist with document processing — tasks that consume significant staff time but do not require judicial reasoning. [3]
Consistency in sentencing and the award of damages is another area where AI may assist. By identifying discrepancies across similar cases, AI tools can help highlight inconsistencies — though the decision itself must always remain a human one.
AI systems are trained on existing data. That data is never neutral. Mauritius has a hybrid legal system — substantive civil law derived from the French Civil Code (Code Civil Mauricien), with procedural and commercial law influenced by English common law. [4] Our jurisprudence inherently contains the societal norms, legal interpretations, and — critically — the biases of the era in which decisions were made. When a machine learning model is trained on this data, there is a real danger that it will replicate those historical biases, embedding old prejudices into seemingly objective analysis.
The question of intellectual independence is paramount. If an AI system provides a statistical success rate of over 90% for a particular outcome, this may exert subtle pressure on the Bench — resulting in a shift from a judiciary driven by equity to one that is, in effect, algorithmically compliant. The judge must always remain the decision-maker; the AI system must remain a tool.
Transparency is equally critical. The phenomenon of the black box — where a system produces a prediction but cannot explain the legal reasoning behind it — is incompatible with the rule of law. A defendant has the right to understand how a decision was reached. If the tool cannot explain itself, it cannot be used in a manner consistent with due process.
Different jurisdictions have responded to the AI challenge in markedly different ways. The experiences of the United States, Europe, and Singapore offer a spectrum of approaches — from aggressive adoption to regulatory-first caution to balanced pragmatism.
| Dimension | 🇺🇸 United States | 🇪🇺 Europe | 🇸🇬 Singapore |
|---|---|---|---|
| **Approach** | Fragmented, state-by-state | Regulatory-first, treaty-based | Balanced, phased, court-led |
| **Key Framework** | COMPAS; state AI committees | CEPEJ Ethical Charter (2018); AI Act (2024) | Registrar's Circular No. 1 of 2024 |
| **Key Tool** | COMPAS risk assessment | European e-Justice Strategy 2024–2028 | Harvey AI (Small Claims Tribunal) |
| **Strength** | Innovation speed, pilot diversity | Ethical framework, binding treaty | Practical deployment + guardrails |
| **Key Risk** | Racial bias, trade secrecy, hallucinated citations | Over-regulation, GDPR tensions | Data confidentiality, mission creep |
| **Maturity** | Operational for decades | Framework-first (evolving) | Early deployment (2024–2025) |
| **Relevance to Mauritius** | Cautionary tale on bias | Model for ethical governance | Model for balanced adoption |
The United States has the longest experience with AI in courts — and the most cautionary lessons to offer.
The COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) system has been used in multiple US states — including New York, Pennsylvania, Wisconsin, California, and Florida — to assess defendant recidivism risk. Judges receive a risk score on a 1–10 scale to inform sentencing decisions. The system processes defendant data (criminal history, age, prior offences) to generate a risk profile, enabling more consistent, data-informed sentencing. [5] · [6]
Standardised risk assessment enabling data-informed sentencing
The National Center for State Courts (NCSC) has documented significant efficiency gains from AI-powered automation in court administration. AI tools assist with data entry in Case Management Systems (CMS), document processing, and e-filing — reducing error rates and freeing court staff for higher-value work. A 2025 NCSC study found that 'greater use of automation in CMS entry could yield major improvements in both efficiency and error rates.' The NCSC published an 'AI Readiness for the State Courts' guide in September 2025. [3] · [7]
Automation of administrative tasks reduces backlog and improves accuracy
By 2024–2025, multiple US states established AI committees and guidelines for courts. Arizona issued Administrative Order 2024-33 creating a Steering Committee on AI and the Courts. Rhode Island's Chief Justice established a Committee on AI and the Courts. Arkansas passed Supreme Court Administrative Order No. 25 on AI (2025). The American Bar Association and state bars issued ethics opinions on AI use in litigation. These frameworks enable courts to adopt AI responsibly while maintaining judicial standards. [8]
Proactive governance frameworks enabling responsible AI adoption
In 2016, ProPublica published a landmark investigation showing that COMPAS was racially biased against Black defendants — falsely flagging them as high-risk at almost twice the rate of white defendants, while white defendants were more likely to be incorrectly flagged as low-risk. The Wisconsin Supreme Court upheld COMPAS use in State v. Loomis (2016–2017), but the case sparked ongoing debate about algorithmic bias in sentencing. [9] · [10]
Algorithmic bias perpetuating racial disparities in sentencing
COMPAS is proprietary software — its methodology is a trade secret. As the Harvard Law Review noted, courts 'cannot evaluate how the risk scores are determined or how the factors are weighed.' This lack of transparency violates due process principles, as defendants cannot challenge the algorithm's logic. Most judges lack the technical expertise to understand or question algorithmic risk assessments. [10]
Proprietary systems prevent judicial scrutiny and defendant challenge
US courts have seen multiple incidents where attorneys submitted legal briefs containing AI-generated fabricated case citations — fake cases invented by generative AI tools. In Mata v. Avianca (2023, SDNY), lawyers were sanctioned for filing a brief with six non-existent cases generated by ChatGPT. This has led multiple federal courts to require AI-use disclosure in filings. [11]
Generative AI fabricates legal citations, undermining court integrity
Europe has taken a regulatory-first approach — building ethical frameworks before deploying AI systems widely in courts.
The Council of Europe's CEPEJ (European Commission for the Efficiency of Justice) adopted the world's first ethical charter on AI in judicial systems in 2018. The Charter provides principles guiding policymakers and justice professionals: respect for human dignity, non-discrimination, transparency, and user control. In 2024, the Council of Europe adopted the Framework Convention on AI and Human Rights, Democracy and the Rule of Law — the first fully binding international AI treaty. [12] · [13]
World's first ethical framework for AI in justice — now a binding treaty
The European e-Justice Strategy 2024–2028, published in the Official Journal of the EU in January 2025, provides a framework for digitalising justice across Member States. The 2025 EU Justice Scoreboard noted that AI assists judges by handling document filing, managing repetitive duties, and researching legislation and case law — allowing judges to focus on substantive legal work. [14] · [15]
EU-wide coordinated digitalisation freeing judges for substantive work
Estonia is considered one of the most efficient court systems in Europe, with automated court procedures and electronic communication tools. The Estonian Ministry of Justice explored AI solutions to reduce administrative burden in courts. While widely reported as a 'robot judge' project for small claims (under €7,000), the Ministry later clarified it does not develop an AI judge to replace human judges — the focus remains on using ICT to reduce court workload and streamline procedures. [16] · [17]
Pioneering e-justice infrastructure reducing administrative burden
The EU's GDPR creates strict requirements for processing personal data — including data used in judicial AI systems. Courts must ensure AI tools comply with data subject rights (right to explanation, right to erasure) while maintaining judicial confidentiality. Eurojust has highlighted tensions between AI-driven judicial cooperation and data protection in criminal matters. The EU AI Act (2024) classifies AI used in justice as 'high-risk,' triggering additional compliance obligations. [18] · [19]
GDPR and AI Act compliance adds complexity to judicial AI deployment
The CEPEJ's revised guidance (2025) explicitly identifies the risk of 'forced use of AI' — where AI systems are integrated in ways that effectively compel judges to rely on algorithmic outputs, violating the 'Under User Control' principle of the CEPEJ Charter. If AI recommendations become de facto mandatory, judicial independence is compromised. The CEPEJ AI Advisory Board flagged this as a key operational risk. [20]
AI integration may subtly coerce judges, undermining independence
In February 2024, a CEPEJ working group published an information note on 'Use of Generative AI by judicial professionals.' The note was notably cautious — described in the International Journal for Court Administration as 'a list of cautions, sometimes impossible to address, that de facto discourage the use of AI applications without considering the benefit that it may bring.' This reflects Europe's tension between innovation and precaution: over-regulation risks stifling beneficial AI adoption in courts. [21]
Over-cautious regulation may stifle beneficial AI adoption
Singapore has taken what observers call a balanced approach: actively piloting AI tools in courts while maintaining strict human oversight. Justice Aidan Xu confirmed judges would try AI 'carefully, and impose appropriate training, safeguards and supervision.' The judiciary issued Registrar's Circular No. 1 of 2024 — one of the world's first court-issued AI guidelines — governing AI use by court users across the Supreme Court, State Courts, and Family Justice Courts. [22]
Singapore's approach is particularly relevant to Mauritius: a jurisdiction with a hybrid legal tradition that adopted AI in courts gradually, with court-issued guidelines first, pilots in low-stakes tribunals, and human oversight throughout.
The Singapore Judiciary signed an MoU with Harvey AI (an American legal startup) to develop an AI-powered platform for the Small Claims Tribunal. The tool summarises case documents for tribunal magistrates (launched September 2025) and was rolled out to self-represented claimants and respondents from November 2025. The tool provides factual summaries without offering case-specific legal advice. Justice Aidan Xu described it as a 'pro bono effort' to improve Access to Justice. [23] · [24]
AI-powered case summarisation improving access to justice for self-represented persons
Singapore issued Registrar's Circular No. 1 of 2024 — 'Guide on the Use of Generative AI Tools by Court Users' — effective 1 October 2024, covering the Supreme Court, State Courts, and Family Justice Courts. The Guide requires court users to verify AI-generated content against authoritative legal databases, use enterprise-grade tools where confidentiality is required, and self-monitor for accuracy. This is one of the first court-issued AI guidelines globally, providing a model for other jurisdictions. [25] · [26]
First-mover regulatory framework balancing innovation with judicial safeguards
In December 2024, the Singapore Judiciary deployed generative AI-powered translation of court documents into Chinese, Malay, and Tamil via a QR code system embedded in Notices of Consultation. This was the initial deployment under the Harvey AI partnership. The tool makes court documents accessible to non-English-speaking litigants in Singapore's multilingual society — directly improving access to justice. [27]
AI translation breaks language barriers in multilingual court access
Singapore's judiciary is acutely aware of AI hallucination risks. The Registrar's Circular explicitly requires court users to verify AI-generated content against authoritative legal databases. Research from the Oxford Institute of Technology and Justice notes that shared concerns across Singapore, South Korea, and Australia include 'hallucinated citations, systemic bias, data privacy, and the need for transparency.' The circular places the burden of verification on the court user. [22]
AI generates fabricated legal citations — verification burden on users
Singapore's guidelines require enterprise-grade AI tools where confidentiality is needed — consumer AI tools risk exposing sensitive case information. The judiciary's Harvey AI partnership uses a closed system, but the risk remains for self-represented persons using public AI tools. The judiciary emphasised that 'case details are securely stored, and confidentiality is maintained at every step.' The challenge is enforcing this standard across all court users. [25]
Ensuring data confidentiality when court users access public AI tools
Justice Aidan Xu noted judicial 'curiosity' and 'enthusiasm' for AI but emphasised it would be deployed 'carefully, and impose appropriate training, safeguards and supervision.' Singapore's approach is deliberately phased: starting with translation, then case summarisation for tribunals, with magistrates first and public access later. The risk is that efficiency gains may create pressure to expand AI use beyond supportive roles into decision-replacement. [22]
Phased adoption must prevent mission creep from support to decision-replacement
The global picture is clear: AI is already in courts worldwide — bringing measurable benefits in efficiency, access, and consistency, but also carrying well-documented risks of bias, opacity, and erosion of judicial independence. No jurisdiction has solved this tension entirely. But some have managed it better than others.
Singapore's balanced approach offers the most relevant model for Mauritius. Both are small jurisdictions with hybrid legal traditions. Both operate in multilingual societies. Both have governments that have made AI a national priority. Singapore started with guidelines first (Registrar's Circular, October 2024), then piloted AI in the lowest-stakes forum (Small Claims Tribunal), and only gradually expanded. Human oversight was never optional.
The transition toward what scholars call Justice 5.0 — human-centred AI in judicial systems — requires the Mauritian judiciary to establish clear guidelines and operational boundaries. [2] The IT Department is no longer solely responsible for understanding these tools. The Bench must also be trained on the mechanics, limitations, and inherent biases of predictive models — so that judicial officers can assess any AI-generated data presented in Court critically and effectively.
Any AI tool used within the Court administration must serve as an aid — supportive, not directive. And it must be transparent. The black box problem — where a system gives a prediction but cannot explain the legal reasoning behind it — is incompatible with the rule of law. The FAIR Guidelines (Fairness, Accountability, Inclusiveness, Integrity, and Responsibility) provide the national ethical framework. The judiciary's own AI guidelines must follow. [28]
UNESCO published its own Guidelines for the Use of AI Systems in Courts and Tribunals in 2025, noting that countries including Argentina, Australia, Brazil, Canada, Colombia, New Zealand, Singapore, the United Kingdom, and the United States had already issued national AI guidelines for courts. [29] Mauritius has the FAIR Guidelines. The next step is judiciary-specific guidance.
Imagine, a decade from now, a Mauritius that has become the legal technology bridge between Africa and the world. The Mauritian judiciary, drawing on its unique hybrid legal system — combining French civil law and English common law — has developed an AI-assisted justice platform that serves as a reference point for African jurisdictions.
Mauritius already ranks third in Africa for ICT development in the ITU 2025 rankings, with a Tier 1 cybersecurity classification. The National AI Strategy (2025–2029) and FAIR Guidelines position Mauritius as one of the few African nations with a comprehensive AI governance framework. Mauritius Telecom's 2026–2029 strategy explicitly targets positioning Mauritius as a digital bridge between Africa and Asia, with four corridors including 'AI & Compute.' [30] · [31]
With its bilingual legal heritage (French and English), its strategic position between African civil-law jurisdictions and Commonwealth common-law systems, and its established FAIR ethical framework, Mauritius could become the natural hub for AI-assisted legal services serving the African continent. Imagine Mauritian-trained AI systems that can analyse case law in both French and English, assist African judiciaries with case management, and provide a transparent, ethically governed AI justice infrastructure that larger jurisdictions — constrained by legacy systems or political fragmentation — struggle to build.
The judiciary that embraces this transformation carefully — learning from the United States' cautionary tales on bias, Europe's ethical frameworks, and Singapore's balanced deployment — could position Mauritius not merely as a consumer of foreign AI legal technology, but as a producer of AI-assisted justice solutions tailored to the African context. The Intelligent Island could become the Intelligent Courthouse for a continent.
The National AI Strategy and FAIR Guidelines have set the stage. But strategy documents do not build systems — people do. So the question is not whether AI will come to the Mauritian judiciary. It is already here, in the smartphones of every lawyer, in the research tools of every chambers, in the case management systems of every registry. The question is whether we will shape it or be shaped by it.
So here are the questions — for the judge, the lawyer, the registrar, the law student, the citizen:
The AI adventure in justice is not something happening to us. It is something we are building together — or failing to build. The National AI Strategy is a roadmap, but every roadmap needs travellers. The judiciary does not need every practitioner to become a technologist. It needs every practitioner to become AI-literate — enough to question, to verify, to demand transparency, and to ensure that AI serves justice, not the other way around.
The question is not whether Mauritius should join this adventure. The question is: what role will you play?