
Johannesburg Labour Court: CCMA AI Hallucination Ruling Overturned
Summary
- A CCMA Commissioner relied on a non-existent judgment, likely generated by AI, to rule on an unfair dismissal claim against SAPO.
- The Johannesburg Labour Court, under Acting Judge Sean Snyman, set aside the Commissioner's ruling, confirming the cited judgment was fake.
- Judge Snyman found the Commissioner failed to verify the AI-generated information, calling it a "fundamental failure of justice."
- The court emphasized that business rescue proceedings generally suspend other legal actions, including unfair dismissal arbitrations.
- This case reinforces a "zero tolerance" stance on unverified AI use in legal research, requiring practitioners to confirm AI outputs.
What Happened
This ruling clarifies that business rescue proceedings generally suspend other legal actions, including unfair dismissal arbitrations, reinforcing the need for rigorous AI legal research verification.
A recent decision by the Johannesburg Labour Court has highlighted critical issues surrounding the use of artificial intelligence in legal research, particularly after a Commission for Conciliation, Mediation and Arbitration (CCMA) commissioner relied on a non-existent judgment. Acting Judge Sean Snyman presided over an urgent application initiated by the business rescue practitioners of the South African Post Office (SAPO). Their objective was to overturn a determination made by Commissioner Menzi Fakude.
Commissioner Fakude had ruled in August 2026 that the ongoing business rescue process for SAPO did not preclude an unfair dismissal claim from proceeding to arbitration. This claim, brought by six former SAPO employees, sought reinstatement and back pay, with arbitration scheduled for October 15. The business rescue practitioners contended that the specific case cited by Commissioner Fakude to support his decision was entirely fabricated. Judge Snyman ultimately sided with the business rescue practitioners on October 15, with his detailed judgment, including the reasons for his decision, being issued late the previous week.
The Legal Framework for Business Rescue
The core of the dispute revolved around the legal implications of a company undergoing business rescue. SAPO had entered business rescue in July 2023, and while its practitioners had applied to the high court in June 2026 to conclude the process, that matter remained pending. The six employees, dismissed in December 2025, had referred their dispute to the CCMA, leading to Commissioner Fakude's August 2026 ruling.
Judge Snyman pointed out that, contrary to established legal principles, the business rescue practitioners had not been included as a party in the CCMA dispute. Furthermore, the Companies Act explicitly prohibits legal proceedings against an entity under business rescue. This statutory provision is designed to safeguard the company's resources and provide it with the necessary stability to achieve commercial viability. Despite these clear legal provisions and established precedent, Commissioner Fakude had concluded that business rescue posed no obstacle to unfair dismissal proceedings, subsequently referring the matter to arbitration.
The Perils of Unverified AI in Legal Research
A central element of Commissioner Fakude's flawed reasoning was his reliance on a purported judgment, S OC Ltd (In Business Rescue) and Others v Communication Workers Union and Others [2023] ZACC 40. He asserted that this case definitively established that business rescue "does not suspend the operation of labour legislation or deprive employees of access to the statutory dispute-resolution mechanisms." However, SAPO's representatives argued that this ruling was based on reasoning contained in a judgment that simply did not exist.
Judge Snyman's own investigation confirmed this assertion; he could find no record of such a case. He discovered that the citation provided by Fakude actually pertained to an unrelated judgment concerning immigration and spousal visas. This discrepancy led Judge Snyman to conclude that Commissioner Fakude "must have relied on AI in seeking to find substantiation for his reasoning." He further criticized Fakude for evidently failing to verify the accuracy of the information, deeming it "an untenable state of affairs" and a "fundamental failure of justice."
A "Zero Tolerance" Stance on AI Hallucinations
The Labour Court's judgment serves as a stern warning regarding the uncritical adoption of artificial intelligence in legal contexts. Judge Snyman highlighted that courts have increasingly expressed concern over legal practitioners' use of AI, referencing a recent full bench appeal where an acting judge was found to have used AI hallucinations. In that instance, Judge Ingrid Opperman advocated for a "zero tolerance" approach to such errors.
Judge Snyman underscored that any arbitrator choosing to employ AI for research is obligated to meticulously verify the existence of any judgment produced by the AI. Furthermore, they must thoroughly read the judgment to confirm that the AI's summary of the decision accurately reflects the actual ruling. He unequivocally declared Commissioner Fakude's ruling to be a "nullity," emphasizing the court's inherent responsibility to intervene in proceedings that are "tainted by illegality." This ruling clarifies that business rescue proceedings generally suspend other legal actions, including unfair dismissal arbitrations, reinforcing the need for rigorous AI legal research verification.
Practical Implications
Lawyers and compliance officers must implement strict protocols for verifying all legal research, particularly when using AI tools, to avoid relying on 'hallucinations' that can lead to rulings being set aside and expose practitioners to professional misconduct. This case underscores the courts' 'zero tolerance' stance on unverified AI use and clarifies that business rescue proceedings generally suspend other legal actions, including unfair dismissal arbitrations.
Source
How does this affect you?
Get an AI analysis of this article grounded in your jurisdictions, practice areas, and any policy documents you've uploaded to Wansom.
Finish Reading the Full Story and the Expert Analysis.
Wansom is AI and can make mistakes.
