“South Africa’s Labour Court intervened in a dispute involving the South African Post Office after a Commission for Conciliation, Mediation and Arbitration (CCMA) ruling cited a Constitutional Court case that did not support the legal proposition attributed to it. The episode has renewed concerns about the use of generative artificial intelligence in legal research and the need for professionals to verify every AI-generated citation against authoritative legal sources.”
South African Labour Court raises alarm over AI-generated legal errors as artificial intelligence reshapes professional work
South Africa’s growing use of artificial intelligence in professional environments has come under renewed scrutiny following a Labour Court judgment involving the South African Post Office (SAPO) and a dispute referred to the Commission for Conciliation, Mediation and Arbitration (CCMA). The case highlights a fundamental challenge associated with generative AI: although these systems can produce convincing explanations and legal references, their outputs may contain fabricated information that can undermine important decisions when users fail to verify the results.
In a judgment delivered on 23 October 2026, Acting Labour Court Judge Sean Snyman criticised a CCMA ruling that relied on a purported Constitutional Court judgment which could not be substantiated as cited. The Labour Court subsequently stayed the arbitration proceedings at the centre of the dispute. MyBroadband reported on the judgment on 27 October.
The decision has implications beyond the parties involved. It illustrates why South African organisations adopting artificial intelligence must establish procedures for checking accuracy, maintaining accountability and ensuring that human professionals remain responsible for consequential decisions.
The legal dispute behind the ruling
The matter concerned six former South African Post Office employees who had been dismissed on 2 December 2025 over alleged misconduct. They approached the CCMA seeking relief, including reinstatement and back pay.
However, SAPO had entered business rescue in 2023. The legal restrictions associated with business rescue created a significant obstacle to proceedings against the organisation. The dispute therefore required consideration of whether the arbitration could proceed in the circumstances.
In addressing this issue, CCMA Commissioner Menzi Fakude relied on legal reasoning that included a purported judgment identified as South African Post Office SOC Ltd (In Business Rescue) and Others v Communication Workers Union and Others, with the citation [2023] ZACC 40.
The citation was challenged because the cited authority could not be found as described. According to the Labour Court judgment, the reference [2023] ZACC 40 actually belonged to Rayment and Others v Minister of Home Affairs and Others, a Constitutional Court matter concerning immigration law rather than the legal question attributed to it in the arbitration ruling.
This distinction matters because legal authorities are not interchangeable. A court or commissioner must rely on judgments that genuinely exist and that support the propositions for which they are cited. A citation that looks authentic but refers to a different case can mislead decision-makers, parties and legal representatives.
How artificial intelligence can produce convincing errors
Generative artificial intelligence systems are designed to produce text, answer questions and identify patterns in information. Large language models generate responses based on learned statistical relationships rather than functioning as infallible databases of verified facts.
As a result, they can sometimes generate nonexistent court cases, incorrect citations, inaccurate quotations or misleading summaries of legislation. These failures are commonly described as AI hallucinations.
The danger is that an invented reference may look entirely plausible. Legal judgments typically follow established naming conventions, use formal language and contain structured citations. An AI-generated citation can reproduce these features convincingly while still being incorrect.
A user who receives such a reference may assume that the system has retrieved an authentic judgment, especially when the answer includes a detailed explanation of the supposed ruling. However, fluency and apparent confidence do not establish factual accuracy.
In the SAPO dispute, the central issue was not simply whether AI had been used. The judgment raised concerns about the failure to establish that the cited legal authority was genuine and relevant before relying on it.
The available reporting indicates that the judge considered AI use a possible explanation for the error. That should not be confused with independently established proof of the exact tool or process used to generate the citation.
Why the Labour Court intervened
Judge Snyman criticised the use of an unsubstantiated legal authority in the arbitration decision. The judgment stressed the importance of diligence and professional responsibility when determining disputes.
The court stayed the arbitration proceedings identified in the judgment and restrained the CCMA and the commissioner from proceeding with them.
The intervention demonstrates that decisions made by administrative and dispute-resolution bodies are not insulated from judicial scrutiny merely because they are made outside ordinary court proceedings.
A dispute-resolution process must still operate within the law. If a decision rests on an unreliable legal foundation, the consequences may extend beyond a simple factual correction. The parties could face delays, additional legal costs and uncertainty over their rights.
For former employees seeking remedies after dismissal, such delays can have practical consequences. For an employer already experiencing financial difficulties, prolonged litigation may also increase the burden of resolving disputes.
The judgment therefore serves as a reminder that accuracy is essential wherever legal reasoning influences the outcome of a case.
Implications for South African legal professionals
Artificial intelligence offers legal professionals several potential benefits. It can assist with document classification, summarising lengthy records, identifying issues for investigation and organising research. Used appropriately, these applications may reduce time spent on routine work and allow professionals to focus on analysis and strategy.
Nevertheless, legal research requires more than producing a plausible answer. Lawyers, commissioners and judges must be able to trace legal propositions to authentic sources and determine whether those sources apply to the facts before them.
A responsible AI-assisted research process should begin with a clearly defined legal question. Any suggested case should then be located in an authoritative database, its full judgment examined, and its relevance checked against the legal issue under consideration.
Researchers should verify the case name, citation, court, date and specific passages relied upon. They should also establish whether the decision remains good law and whether subsequent judgments have distinguished, limited or overturned it.
Importantly, merely finding a judgment with a similar name is not enough. The actual text must support the proposition being advanced.
These safeguards are especially important when AI systems summarise multiple cases or produce a chain of legal reasoning. A single fabricated authority can undermine an otherwise plausible argument, particularly if subsequent conclusions depend on it.
A wider challenge for AI governance in South Africa
The legal controversy arrives as South Africa continues to develop its approach to artificial intelligence governance.
The South African Bureau of Standards has proposed draft AI standards addressing transparency, data quality, human oversight, and the benefits and environmental impacts of AI systems. Public comments are due by 10 November 2026, according to MyBroadband. The proposals are voluntary drafts, rather than a blanket legal requirement for every organisation to adopt the standards.
These issues are directly relevant to professional AI use.
Transparency involves making it possible to understand how an AI system is being used and what information it provides. Data quality concerns the reliability and suitability of the information used in an AI system or supplied to it. Human oversight concerns whether people can monitor, assess and intervene in automated processes.
In legal work, these principles can translate into practical measures: documenting the use of AI, maintaining reliable research records, requiring verification of citations and assigning responsibility to a qualified person who reviews the final work.
Such measures can reduce the likelihood that a generated error will pass unnoticed into an official ruling, legal submission or professional report.
The standards process is also distinct from the development of South Africa’s national AI policy. The earlier draft national policy was withdrawn after problems involving fabricated references were identified. A revised policy process is underway, according to previous reporting by TimesLIVE and Reuters.
The wider policy challenge is to encourage innovation while ensuring that AI use does not compromise public trust or institutional accountability.
What organisations should do before adopting AI
The lesson for organisations is not that every application of artificial intelligence should be abandoned. Rather, the level of verification should match the importance of the task.
For low-risk administrative work, an organisation may require routine checking for accuracy and appropriateness. For legal, medical, financial or other consequential work, the verification process should be significantly more rigorous.
Organisations using AI for legal research should establish approved tools, clear usage policies and training requirements. Employees should understand that generated text is a starting point for investigation, not automatically a reliable source of evidence.
Where AI contributes to an important decision, the organisation should also maintain a record of the sources consulted, the checks performed and the person who approved the final output.
Training is particularly important because the ability to use AI effectively involves more than writing prompts. Users must recognise the technology’s limitations and know how to challenge its answers.
Managers should avoid measuring AI success solely by speed or the volume of documents produced. Accuracy, reliability and the cost of correcting errors should also inform decisions about whether an AI application is genuinely useful.
Protecting public confidence in an AI-driven future
The Labour Court dispute illustrates a central tension in the adoption of artificial intelligence. The technology can make information easier to generate, but it cannot eliminate the responsibility to establish whether that information is true.
This is especially important in legal proceedings, where the integrity of decisions depends on accurate facts, applicable law and fair procedures. A fabricated citation can damage the credibility of a decision even when other parts of the reasoning appear persuasive.
For South Africa, the challenge is to develop an environment in which innovation and accountability reinforce one another. Businesses, government departments and professional bodies can benefit from AI, provided they invest in appropriate safeguards and do not confuse automated output with verified knowledge.
The case also reinforces the continuing importance of authoritative legal databases, professional judgement and independent review. AI may help locate potential arguments, summarise material and organise research, but the responsibility for relying on a legal authority remains with the human decision-maker.
Ultimately, the value of artificial intelligence will depend not only on how quickly it can generate answers, but also on whether people can trust the processes through which those answers are checked and used. The South African Labour Court’s intervention is a timely reminder that professional accountability must remain central as AI becomes more deeply integrated into the country’s institutions.
