Into this landscape arrives artificial intelligence (AI): a technology that proponents herald as capable of democratising access to legal information and services. The potential is considerable given that AI tools can now draft legal documents, summarise complex case materials, conduct preliminary legal research, and provide guidance on procedural matters at a fraction of traditional costs. For a jurisdiction where a significant proportion of litigants appear without legal representation, these capabilities could prove transformative.

Yet the integration of AI into justice systems raises fundamental questions that merit careful examination. Can algorithmic decision-making satisfy constitutional requirements of fairness? How do we ensure that technological efficiency does not compromise the essential human elements of adjudication? Can AI systems that are not secured compromise the legal privilege afforded to individuals through common law principles? What safeguards are necessary to prevent AI systems from perpetuating or amplifying existing biases? These questions are not merely academic; they will shape whether AI serves as a bridge to justice or creates new barriers in its stead.

For a detailed examination of AI’s implications for legal professional privilege, see our previous Alert.

The global picture: AI enters the courtroom

The question of whether machines should decide cases is no longer hypothetical. Across the globe, judicial systems are grappling with AI’s expanding role. Estonia is piloting AI adjudication of small claims disputes below EUR 7,000. China has declared the integration of AI into judicial processes a national priority, deploying precedent recommendation systems that formulate judgments based on prior decisions. In the US, algorithmic risk assessment tools already inform bail and sentencing decisions in numerous jurisdictions, though typically in an advisory rather than determinative capacity.

The UK offers a particularly instructive comparative reference. In October 2025, the UK judiciary issued updated guidance on AI use by judicial office holders, establishing principles for responsible deployment. The guidance emphasises that any use of AI must be consistent with the judiciary’s overarching obligation to protect the integrity of the administration of justice. It cautions that publicly available AI chatbots do not provide answers from authoritative databases but rather generate text based on statistical predictions. This text may be inaccurate, incomplete, misleading or biased. Judicial office holders are reminded that they remain personally responsible for material produced in their name, and that AI tools cannot replace direct judicial engagement with evidence.

This cautious approach reflects broader concerns about AI’s limitations. Academic research has demonstrated a perceived “human-AI fairness gap”: experimental studies show that ordinary citizens evaluate AI-led judicial proceedings as less procedurally fair than those conducted by human judges. This gap persists across different legal contexts, from consumer arbitration to criminal sentencing. Importantly, however, the research also indicates that this fairness gap is neither irreducible nor absolute; it can be mitigated through what scholars term “algorithmic offsetting” by providing litigants meaningful hearings and ensuring AI decisions are interpretable and transparent.

Opportunities for South Africa

For South Africa, AI presents a compelling opportunity to address entrenched access to justice barriers. The technology’s most immediate promise lies in its capacity to reduce the cost of legal services. Scholars have noted that AI tools sharply reduce the costs of generating legal materials, potentially allowing many more people to access justice. Legal sociologists identify multiple barriers that prevent individuals from pursuing legitimate claims. These include not only the cost of lawyers but also difficulties in recognising that a grievance constitutes a legal wrong, understanding available remedies and navigating procedural requirements. AI can assist at each of these stages.

Consider the practical applications already emerging. In the arbitration context, the Association of Arbitrators (Southern Africa) issued AI guidelines in May 2025, recognising that AI tools are already embedded in South African arbitration proceedings. These tools assist with collating and sequencing complex case facts and chronologies, managing documents and expediting the reviewing of large volumes of content, conducting legal research and sourcing precedents, drafting submissions and procedural documents, and facilitating real-time translation or transcription during hearings.

The potential efficiency gains are substantial. One analysis suggests that AI could enable legal aid organisations to serve significantly more clients with existing resources, with some projections indicating capacity increases of 26% to 50% or more. For a country where legal aid resources are stretched thin, such gains could meaningfully expand access to representation. South Africa’s investment in AI infrastructure, including the establishment of Africa’s first operational AI factory and extensive data centre capacity, positions the country to develop locally contextualised legal AI solutions.

Integration into judicial administration offers particular promise. AI systems can assist with document summarisation, allowing judges to efficiently process voluminous case materials. They can support case management through intelligent scheduling and prioritisation. For overburdened courts, such tools could help address backlogs without compromising the quality of individual case consideration. As one scholar observes, the goal should be to leverage technology to scale up and improve the delivery of justice without sacrificing justice in individual cases.

The risks: Bias, transparency, and the human element

The enthusiasm for AI’s potential must be tempered by a clear-eyed assessment of its limitations. The risks are neither speculative nor merely theoretical, they are already manifesting in courtrooms across the world, including in South Africa.

The phenomenon of “hallucinations”, where AI systems generate plausible-sounding but entirely fabricated information, including fictitious case citations, has emerged as a significant concern. South African courts have already encountered submissions containing AI-generated fictitious legal authorities.
In a June 2025 matter involving Northbound Processing and the South African Diamond and Precious Metals Regulatory Authority, AI-generated errors were identified in court documents. Similar incidents have occurred globally, prompting courts to impose sanctions on legal practitioners who failed to verify AI outputs. The Legal Practice Council has begun receiving referrals concerning practitioners’ use of AI in preparing court materials.

Algorithmic bias presents a more insidious challenge. AI systems are trained on historical data, and that data inevitably reflects the biases and inequities embedded in past human decisions. As one commentator has noted, the risk is “bias in, bias out”. In a jurisdiction still confronting the legacy of systemic discrimination, this concern takes on particular urgency. An AI system trained predominantly on data from other jurisdictions may embed assumptions and legal frameworks ill-suited to South Africa’s constitutional order and societal context.

Perhaps most fundamentally, scholars have articulated concerns about a “right to a human decision”. This argument posits that certain decisions affecting individual rights and liberties should be made by human beings capable of moral reasoning, empathy, and accountability. As Professor Aziz Huq has argued, there are contexts where human judgement possesses qualities that cannot be replicated algorithmically. The constitutional right to a fair hearing arguably encompasses not merely formal procedural compliance but substantive engagement by a decision-maker capable of genuinely understanding the litigant’s position.

Experimental research provides empirical grounding for these concerns. Studies demonstrate that people generally perceive human judges as procedurally fairer than AI judges. This perception matters given that research in legal psychology establishes a relationship between perceived fairness and legal compliance. If citizens regard AI-assisted decisions as less legitimate, they may be less inclined to accept and comply with those decisions, potentially undermining the very objectives the technology is meant to serve.

The regulatory landscape

South Africa’s regulatory framework for AI remains nascent but is evolving. The draft National Artificial Intelligence Policy Framework, developed by the Department of Communications and Digital Technologies, completed its consultation phase in 2025, proceeded to Cabinet, and was published for public comment on 10 April 2026. However, it was withdrawn on 26 April 2026 following the discovery of fabricated citations in its reference list, an ironic illustration of the very risks AI outputs can pose. Once finalised, the framework is intended to align with international standards, drawing on the United Nations Educational, Scientific and Cultural Organization’s Recommendation on the Ethics of AI, the Organisation for Economic Co-operation and Development’s AI Principles, and the Council of Europe’s AI Treaty. It emphasises fairness, transparency, accountability, privacy, safety, human oversight, and cultural values.

A distinctive feature of South Africa’s approach is its emphasis on local cultural and societal values as standards for AI regulation, a recognition that AI systems developed without consideration for African contexts and priorities may exacerbate existing inequalities rather than remediate them. The policy acknowledges that when it comes to AI regulation, there is no one-size-fits-all approach and African states will need to tailor-make their AI legal frameworks to fit each country’s specific context.

At present, AI in South Africa is primarily regulated through existing data privacy legislation, namely the Protection of Personal Information Act 4 of 2013 (POPIA). While POPIA provides some guardrails, particularly concerning the collection and processing of personal data, it was not designed to address the full spectrum of AI-specific risks. It does not comprehensively cover the AI lifecycle from development to deployment, nor does it address concerns such as algorithmic transparency or explainability that are increasingly central to responsible AI governance.

Legal practitioners operating in this environment must navigate a patchwork of applicable laws, including POPIA, the Electronic Communications and Transactions Act 25 of 2002, and professional conduct rules. In the absence of dedicated AI legislation, practitioners bear significant responsibility for ensuring that AI tools are used ethically and that outputs are independently verified. The legal profession’s existing duties of competence and diligence apply with equal force to AI-assisted work.

Recommendations

For clients and business leaders	For legal practitioners	For policymakers and the judiciary
Before deploying AI in legal and compliance functions, organisations should establish governance frameworks that include clear usage policies, human oversight requirements and verification protocols to prevent errors. Where personal information is involved, POPIA obligations apply, and vendor contracts should address data sovereignty, confidentiality and liability.	Fictitious AI-generated citations in South African courts highlight the need for independent verification of all AI outputs. Every case, citation and legal proposition must be confirmed. AI is a secondary tool, not a substitute for professional judgement. Practitioners should understand AI’s limitations, remain accountable for all material produced and never enter confidential information into public AI systems.	Proactive AI integration is preferable to reactive adjustments that historically narrow rights or raise procedural barriers. The judicial system should pilot AI tools in controlled settings, develop guidance similar to the UK’s approach and train judicial officers and court staff. The aim is to enhance judicial capacity while preserving the quality of case consideration and the human elements of adjudication.

The integration of artificial intelligence into South Africa’s justice system is not a question of whether, but how and when. The technology is already here and it is being used by legal practitioners, arbitrators, and increasingly by members of the public seeking to understand their rights. The question before policymakers, the judiciary, and the legal profession is whether AI’s deployment will be guided by thoughtful planning or left to ad hoc adoption with all its attendant risks.

The stakes are high. AI offers genuine potential to democratise access to legal services and extend the reach of justice to those currently excluded by cost, geography or information barriers. Yet that potential will be realised only if deployment is accompanied by robust safeguards against bias, meaningful human oversight and transparency in algorithmic decision-making.

The use of AI in arbitration and dispute resolution may support justice, but it cannot replace those who are tasked with safeguarding it.

South Africa stands at a crossroads. The choices made in the coming years in developing AI policy, establishing judicial guidance, training legal practitioners and educating the public will shape whether AI serves as an instrument of greater justice or introduces new forms of exclusion and error. What we gain in access to justice we must not lose in the delivery of justice. The path forward demands both ambition and caution: ambition to harness technology in service of constitutional ideals, and caution to preserve the human elements that give those ideals meaning.