The South African Bureau of Standards (SABS) has published a series of draft standards on artificial intelligence (AI) for public comment.

The drafts adopt international ISO/IEC standards as South African documents, and cover transparency, data quality, human oversight, and the benefits and environmental impact of AI. While the standards are voluntary, they are likely to become the local benchmark for responsible AI, in line with international best practice, creating alignment between South Africa and global standards.

Interested parties, including organisations that develop or use AI, may submit comments by 10 November 2026.

The SABS has published draft standards, technical specifications and technical reports governing the use of AI in South Africa.

The drafts were prepared by the SABS Technical Committee on Information Technology and Artificial Intelligence (SABS/TC 001/SC 42), which has proposed that the documents should become South African standards, technical reports and technical specifications.

Each draft is an identical adoption of an existing ISO/IEC document on artificial intelligence, without local changes. In other words, the international text is replicated in the South African documents.

The International Organization for Standardization (ISO) and the International Electrotechnical Commission (IEC) are independent international bodies that develop standards through experts nominated by ISO member bodies. Their joint AI committee is ISO/IEC JTC 1/SC 42, which the local SABS committee mirrors. Because ISO/IEC standards are developed by consensus across many countries, they are widely recognised by regulators, investors and business partners, and are often used as a reference point for good practice.

The identical adoption of ISO/IEC standards is a worldwide norm. It promotes conformity with international benchmarks used by trading partners, and enables smaller countries to benefit from the work done by experts from many countries, rather than each country writing its own standards, promoting international consistency and certainty.

The drafts seek to create voluntary standards and specifications for organisations deploying AI, in the following four areas:

• Transparency (SANS 12792 / ISO/IEC 12792:2025): This sets out a common structure for the information that should be disclosed about an AI system, covering the system itself, the underlying model, the datasets used and the wider societal and environmental context. The aim is to help developers, deployers, auditors, authorities and end users understand how an AI system works, what it is meant to do and what its limits are.
• Data quality (SANS 2620 Parts 1, 2, 3 and 5 / ISO/IEC 5259-1, -2, -3 and -5): Data quality is critical to reliable AI, because AI systems learn from data sets. These parts set out key concepts and the data life cycle, measures for data quality characteristics such as accuracy, completeness and representativeness, requirements for managing data quality across the life cycle, and a governance framework for boards and management to direct and oversee data quality.
• Human oversight (SATS 8200 / ISO/IEC TS 8200:2024): This deals with the controllability of automated AI systems. It sets out a framework so that a human or other controller can monitor the state of an AI system and take over control when needed, with different levels of controllability and steps to test that these controls work.
• Benefit and environmental impact (SATR 24030 / ISO/IEC TR 24030:2024 and SATR 20226 / ISO/IEC TR 20226:2025):
  • The use case report collects 81 real-world AI use case summaries across sectors such as fintech, e-commerce, energy, agriculture, education, healthcare, insurance, manufacturing, media and entertainment, security and transport.
  • The environmental report gives an overview of the environmental footprint of AI systems across their life cycle and supply chain, including energy use, carbon, water and e-waste, and maps possible metrics and actions to reduce that impact.

Such standards are not legally binding (unless they have been made compulsory, for example by being referenced in law). Rather, they create voluntary standards which organisations can choose to adopt to improve quality, safety, transparency and the like, in accordance with the standards developed by internationally recognised experts.

How to comment

Anyone can comment on the draft standards by 10 November 2026, by email to dsscomments@sabs.co.za.

Comments must be submitted using the SABS commenting template (SABS_commenting_template.doc), with reference to the SABS User Guide.

The post SABS publishes draft AI standards for public comment appeared first on Werksmans Attorneys.

The Labour Court’s decision in Bosch v Passenger Rail Agency of SA provides employers with a useful warning: when designing voluntary severance or early-retirement programmes, age may be an administratively convenient dividing line, but convenience does not make discrimination lawful. The dispute arose from PRASA’s efforts to reduce its wage bill and headcount. It introduced a voluntary severance package (“VSP”) for employees generally, while employees between the ages of 60 and 63 were instead offered voluntary early retirement (“VER”) where the VSP would have produced a higher amount. The financial distinction was significant. Employees younger than 60 could receive a VSP incorporating a “sweetener” of 2.5 weeks’ salary for every completed year of service. Employees between 60 and 63 were limited to a lump sum calculated according to the months remaining until retirement.

Bosch was 62 and had more than 44 years’ service. He initially obtained a VSP quote exceeding R2 million and submitted his application. He was subsequently informed that his application had been treated as one for early retirement and that he would receive R657,906.67 – effectively seven months’ salary plus notice pay.  After unsuccessfully pursuing PRASA’s internal grievance process, Bosch challenged the distinction as unfair discrimination based on age under section 6(1) of the Employment Equity Act (“EEA”). The difficulty for PRASA was straightforward: the different treatment was expressly determined by age. The Court found that the only reason Bosch received the substantially lower VER package rather than the VSP was that he was over 60.

Once discrimination on a listed ground such as age is established, section 11(1) of the EEA places the burden on the employer to prove that the discrimination did not occur, or that it was rational and not unfair, or otherwise justifiable. PRASA struggled at this crucial stage. Its witnesses had not made the decision to differentiate between employees according to age and could not explain why the board had done so. No board member, or anyone else involved in the decision, testified. The Court was therefore left without admissible evidence explaining why the distinction had been adopted or demonstrating the savings supposedly achieved by it.

PRASA also argued that Bosch could have withdrawn his application once he knew what he would receive and that, by accepting the VER payment, he had effectively waived his right to complain. The Court rejected this argument. It held that Bosch’s acceptance of the payment did not waive his right to challenge the discriminatory treatment. PRASA’s reliance on a “full and final settlement” provision similarly could not rescue the differentiation. This is an important practical point for employers. A signed separation document should not be treated as an automatic cure for an underlying discrimination problem.

The Court ultimately held that PRASA’s policy breached the EEA and that the failure to pay Bosch the VSP constituted unfair age discrimination. PRASA was ordered to pay a global amount of approximately R1.59 million, subject to tax, together with interest and costs. The judgment does not mean that employers can never differentiate between employees when designing restructuring, severance or retirement arrangements. It does, however, emphasise that where the differentiation is directly linked to a listed ground such as age, the employer must be able to justify both the distinction and the rationale underlying it with evidence. For employers contemplating voluntary retrenchment or early-retirement programmes, that justification should therefore be interrogated – and documented – when the scheme is designed, rather than reconstructed when the matter reaches court.

The post Too old for the package? Labour Court draws the line on age-based severance appeared first on Werksmans Attorneys.

Introduction

The Gauteng Division of the High Court of South Africa recently published a notice of proposed amendments (“Proposed Amendments“) to its Mandatory Mediation Protocol (“Mediation Protocol”), which operationalises the Division’s Mandatory Mediation Directive. The proposed amendments are an attempt to reduce the clogging in the Special Interlocutory Court (“SIC“).

The Purpose of the SIC

The SIC was established to address procedural non-compliance by an adversary in litigation. The Division’s practice directive states that “the very purpose of the SIC is to overcome delinquency and consequent improper delays”. It further provides that any litigant who is aggrieved by another litigant’s “neglect, dilatoriness, failure, or refusal to comply” with any provision of the Uniform Rules of Court (“Rules“), practice manual, or the practice directive must approach the SIC to compel compliance from the delinquent party.

Notably, the orders obtainable in the SIC are strictly of an interlocutory nature and therefore do not deal with the substantive (or main) aspects of the litigation. By having a dedicated court of this nature, it was envisaged that there would be a reduction of the time spent on interlocutory issues which would create more capacity for other court rolls to deal with substantive issues.

Unfortunately, despite best intentions, the SIC as an innovation has faced a significant functional problem – excessive congestion. The Proposed Amendments highlight that at any given time there are approximately 16 000 matters seeking allocation and hearing dates in the SIC, which causes congestion, delays and backlogs.

In some ways, this congestion is a self-fulfilling prophecy. Mandatory Mediation, pursuant to the procedures in the Mediation Protocol, was meant to free up the court rolls and foster the expedient and negotiated settlement of cases between parties. Instead, the Mediation Protocol simply moved the traffic from one intersection to another, and the SIC suffers for it. Resultantly, despite the relative availability of court dates, the cases that previously clogged up the court rolls did not disappear, they have now just clogged up the system meant to expediently resolve procedural issues. Accordingly, the proposed amendments are necessitated by the SIC backlog of procedural delinquency cases, owing to parties not fully complying with the Mediation Protocol.

The Mediation Protocol

The introduction of mandatory mediation in Gauteng, accompanied by the Mediation Protocol to operationalise the system, also made peremptory the rules of procedure pertaining to mediation.

From the perspective of the Gauteng Division, mediation was meant to promote alternative dispute resolution and ensure that only those cases which absolutely require judicial intervention come before the judges of the Division. However, alternative dispute resolution processes require the active participation of all parties to a dispute, failing which the peremptory mediation rules are undermined. Accordingly, and as an example, where a party fails to respond to a notice calling for mediation, that constitutes non-compliance and such delinquency may be brought before the SIC for what should ideally be expedient resolution. The Proposed Amendments explain that the SIC congestion is primarily due to litigants’ non-responsiveness to Rule 41A mediation notices. This not only undermines the underlying rationale of the SIC which, in addition to what is stated above, includes the prevention of delay and gamesmanship but also the effective implementation of the Mediation Protocol.

In Malebane v Road Accident Fund [2025] ZAGPPHC 1253, the court remarked that of approximately 200 applications serving before the SIC during that week, a significant number were “Delinquency Applications”. The court further noted that most, if not all, of these applications arose from respondents’ failures to respond to Rule 41A mediation notices. The court expressed growing concern that delinquency applications were increasingly being used as tactical shortcuts to gain access to the Default Judgment Roll. This is an example of the exact gamesmanship the SIC is meant to prevent. The court’s observations are particularly relevant to the proposed amendments. The fact that the majority of matters before the SIC comprise of applications of this nature underscores the need for measures aimed at addressing their increase and lessening the misuse of the delinquency mechanism. The proposed amendments are an attempt to address this issue.

The Proposed Amendments

The solution suggested in the Proposed Amendments is to excise matters where the only issue is a litigant’s failure to reply to a Rule 41A mediation notice. In terms of the Proposed Amendments, mediation delinquency would instead be dealt with through the procedure prescribed in Rule 30A of the Rules which governs non‑compliance with the Rules and Court Orders. This marks a departure from the existing procedure in the Mediation Protocol, which requires an aggrieved party to obtain an order from the SIC compelling the delinquent party to comply with the Mediation Protocol, failing which the aggrieved party may seek that: the delinquent litigant’s claim or defence be struck out, the matter be enrolled on the Default Judgment Roll, and the delinquent litigant be held to be in contempt of the compelling order.

In terms of the proposed solution where a party fails to deliver a response to a Rule 41A mediation notice, the aggrieved party (who delivered such notice) may serve a Rule 30A Notice calling upon the delinquent party to respond within 10 days of delivery of that notice, failing which the aggrieved party may apply to the relevant Registrar for the allocation of the matter on the Default Judgment Roll or the Special Default Judgment Roll if the delinquent party fails to deliver a response after 15 days of expiry of the period called for in the Rule 30A Notice.

Conclusion

While the proposed amendments may result in a reduction of the congestion in the SIC and support the effective implementation of the mandatory mediation regime, they may also just result in another unsuccessful re-direction of judicial traffic; this time from the SIC to the Default Judgment Roll and the Special Default Judgment Roll. It is contemplated in the Proposed Amendments that if the proposed solutions are approved, then transitional provisions to facilitate their pragmatic implementation should be developed and adopted.

It is hoped that the proposed changes have the desired effect and improve the administration of justice in the Gauteng Division. The importance of effective judicial administration cannot be understated, especially given the Gauteng Division’s place as the busiest court in the country.

The post Proposed Amendments to the Mandatory Mediation Protocol in the Gauteng Division of the High Court: Will the latest attempt to redirect judicial traffic succeed? appeared first on Werksmans Attorneys.

1. Section 11 of the Mineral and Petroleum Resources Development Act 28 of 2002 (“the MPRDA”) is one of the most frequently invoked provisions in the mining regulatory landscape. Every cession, transfer or disposal of a prospecting or mining right, or of an interest in such a right is subject to its provisions. It is, however, easy to lose sight of what the consent process is actually designed to do. A recent judgment of the High Court of South Africa, Gauteng Division, Pretoria, serves as a timely reminder that ministerial consent is a regulatory approval of a transaction, not a cure for a transaction that never lawfully came into existence.

2. In Misty Falls 45 (Pty) Ltd and Another v The Minister of Mineral Resources and Energy and Others (Case No. 19058/2022) (“the Misty Falls Judgment”), handed down by Raubenheimer AJ and deemed delivered on 10 April 2026, the Court reviewed and set aside the Director-General’s (“DG”) consent to the transfer of the Kareepan mining right, as well as the Minister’s dismissal of the internal appeal against that consent. Unusually, the Court went further and substituted its own decision: consent was refused.

3. By way of background, Misty Falls 45 (Pty) Ltd (“Misty Falls 45”) holds a manganese and iron ore mining right over portions of the farms Kareepan 450 and Pensfontein 449 in the Kuruman district. A former director and shareholder (“the former director”) sought, together with a director of Misty Falls 111 (Pty) Ltd (“Misty Falls 111”), consent under section 11 to transfer that right to Misty Falls 111.

4. The former director relied on a 2010 “Exchange of Rights Agreement” between shareholders and a 2011 shareholders’ resolution as the source of his authority. The documentary record, however, showed that he had resigned as director in 2014. No special resolution in terms of sections 112 and 115 of the Companies Act 71 of 2008 was ever passed, and the then board expressly told the Department that no transaction involving the right should be processed without its permission.

5. The underlying causa also shifted over time. A February 2018 letter referred to a “cession agreement” between Misty Falls 45 and Misty Falls 111 which did not exist. A 2019 motivation letter relied instead on a November 2018 agreement between the former director personally and Misty Falls 111.

6. The DG nonetheless granted consent in January 2020, focusing on the transferee’s financial and technical capacity without engaging with the questions of authority and causa. On appeal, the Department’s Chief Director: Legal Services identified authority as the “crisp issue” and recommended that the appeal be upheld. The Minister dismissed the appeal two days later without engaging with that recommendation.

7. Section 11(1) provides that a prospecting or mining right, or an interest in it, may not be ceded, transferred, let, sub-let, assigned, alienated or otherwise disposed of without the Minister’s written consent. Section 11(2) provides that consent “must be granted” if the transferee can comply with the obligations and conditions of the right and satisfies the requirements of section 17 or section 23, as the case may be. On its face, therefore, the enquiry is directed at the transferee and its capacity, and in practice that is where the attention of the Department is usually focused.

8. The Court, however, identified a logically prior question. As Raubenheimer AJ put it, section 11 “presupposes the existence of a lawful cession, transfer, alienation or disposal of a right”, and the Minister’s consent “cannot validate an otherwise unlawful or unauthorised transaction”. Put simply, there must first be a valid transaction to which consent can attach.

9. Applying that principle, the Court found that none of the sources of authority relied upon could sustain the transfer-

9.1. The Exchange of Rights Agreement dealt only with the allocation of economic benefits from two prospecting rights among shareholders. Section 19(1)(b) confers on the holder of the prospecting right the exclusive right to apply for a mining right – this right is not conferred on individual shareholders. Reading the agreement as transferring a future mining right would render it unlawful as an attempt to circumvent section 11.

9.2. The 2011 resolution, properly construed, was aimed at establishing and operating the mine, not at disposing of the right itself. In any event, a private resolution cannot override the peremptory requirements of sections 112 and 115 of the Companies Act.

9.3. Whatever authority the former director held as a director lapsed upon his resignation, and a shareholder, even a majority shareholder, has no inherent authority to bind a company to the disposal of its assets.

10. The review Court accordingly concluded that the section 11 application was neither authorised by Misty Falls 45 nor supported by a valid underlying transaction, and that the consent was not authorised by the empowering provision. In considering remedy, the Court described the “central defect” as “the absence of both authority and a valid underlying transaction” and held that there was “no conceivable basis” on which the DG or the Minister could lawfully grant consent on the facts. Refusal was the only lawful outcome, and remittal “would serve no purpose other than to delay the inevitable”.

11. The Court also addressed the limits of the regulator’s role. The Minister is not empowered to adjudicate private disputes about shareholding, directorship or internal corporate arrangements, save to the extent necessary to determine whether the application is duly authorised and lawful. Faced with manifest disputes about authority, the Minister was required either to refuse consent or to insist that the parties resolve their dispute in the appropriate forum. By “siding” with one faction on contested questions of corporate law, the Minister exceeded the scope of his section 11 powers. In the Court’s words, the regulator’s role “is not to confer legitimacy on contested private arrangements that may themselves be unlawful or ultra vires”.

12. The significance of the Misty Falls Judgment lies in its clear statement that a void transaction cannot be rescued through section 11. Consent is not a form of administrative ratification. The Court was careful to note that nothing prevents the parties from structuring a lawful, properly authorised transaction in future and applying afresh, but what is impermissible is “to validate, through administrative fiat, an unauthorised and opaque transaction”. The Court also held that, while the capacity assessment under section 23(1)(b) attracts a more deferential standard, decision-makers remain obliged to consider whether the underlying transaction and authority are sound.

13. For the industry, the implications are practical. Regulators processing section 11 applications cannot confine themselves to a box-ticking assessment of the transferee’s financial and technical capacity, particularly where objections from the right holder’s board are on record. Practitioners should ensure that applications are supported by a written transaction concluded by the right holder itself, proper board authority and, where the right constitutes all or the greater part of the company’s assets, the requisite special resolutions. Market participants acquiring rights should treat the existence of internal disputes as a serious red flag, since a consent obtained on a defective foundation is vulnerable to being set aside, with adverse costs to follow.

14. Whilst the Misty Falls Judgment is binding only in the Gauteng Division, it is persuasive elsewhere and restates a principle of general application: section 11 regulates lawful transactions, it does not create them. Where the foundation is absent, there is simply nothing for the regulator to consent to.

The post Section 11 consent: Can the regulator breathe life into a void transaction? appeared first on Werksmans Attorneys.

Constructive dismissal is not an escape route for employees who are dissatisfied with their working environment, nor does every workplace grievance entitle an employee to resign and demand compensation. The Labour Appeal Court’s recent judgment in Royale Energy (Pty) Ltd v Commission for Conciliation, Mediation & Arbitration & others (2026) 47 ILJ 2419 (LAC) provides a welcome reminder that intolerability is a high threshold, and that employers must be afforded a reasonable opportunity to address workplace complaints before employees resort to resignation.

The employee, a senior manager at Royale Energy, had experienced a difficult relationship with the company’s group chief executive officer over several years. Following an incident in August 2020, during which the CEO allegedly behaved aggressively and threateningly towards her, she lodged a formal grievance and requested the board’s intervention.  The employer responded by appointing an external investigator and proposing alternative reporting arrangements while the investigation continued. The employee, however, rejected the proposed arrangements, sought a mutual separation package exceeding R4 million and, when this was declined, demanded the investigation report.

Less than a month after lodging her grievance, and before the investigation had been completed, she resigned and referred a constructive dismissal dispute to the CCMA.  The commissioner found in her favour, a decision subsequently upheld by the Labour Court, which awarded compensation of approximately R1.15 million.

The Labour Appeal Court took a very different view and held that intolerability is not simply unhappiness. Section 186(1)(e) of the Labour Relations Act provides that constructive dismissal occurs where an employee terminates employment because the employer has made continued employment intolerable.  The Court reaffirmed that this requires an objective assessment. It is insufficient for an employee merely to believe that the workplace has become unbearable. The employer’s conduct must objectively render continued employment intolerable.

Importantly, the Court distinguished between an unpleasant or difficult working relationship and circumstances genuinely leaving an employee with no reasonable alternative but to resign. The employee had continued working despite several earlier incidents involving the CEO, none of which had been formally escalated to the board. When she eventually complained, the employer took steps to address her concerns. These facts were inconsistent with her contention that continued employment had become objectively intolerable.

Perhaps the most significant aspect of the judgment is the Court’s treatment of the employer’s response to the grievance. The employee had expressly requested intervention and indicated her willingness to participate in a process aimed at resolving the dispute. Yet, when the employer initiated precisely such a process, she resigned before allowing it to run its course.  The Court was particularly critical of the employee’s attempt to use the ongoing remedial process itself as evidence of intolerability. An employer cannot reasonably be expected to investigate allegations spanning several years, consider the competing versions and implement appropriate corrective measures within an arbitrarily short period.  Nor can an employee invoke a grievance procedure, refuse to allow it to reach its conclusion, and then rely on the absence of an immediate resolution as justification for resignation.

The Court also rejected the employee’s reliance on anticipated future misconduct. Constructive dismissal cannot be founded on speculation that working conditions might become intolerable at some future point – the employer must be given a fair opportunity to resolve the complaint.  Employers are entitled to investigate complaints, implement interim protective measures and explore workable solutions without automatically exposing themselves to constructive dismissal claims. This does not excuse genuine workplace bullying or permit employers to ignore legitimate grievances. It does, however, recognise that a properly constituted remedial process requires time and cooperation. The Labour Appeal Court found that no constructive dismissal had occurred.

For employers, the message is clear. A prompt, documented and reasonable response to workplace grievances can provide a powerful defence against constructive dismissal claims.  For employees, the lesson is equally important: a resignation does not become a dismissal merely because the employee is unhappy with the employer’s response.  The law protects employees against genuinely intolerable working conditions. It does not guarantee an immediate or preferred outcome to every workplace complaint.

The post Resign First, Complain Later? The LAC Draws the Line on Constructive Dismissal appeared first on Werksmans Attorneys.

Environmental authorisations impose strict obligations which continue throughout the life of a project. In instances where unlawful pollution persists, preparing remedial plans does not discharge those obligations, and regulators cannot remain inactive in the face of non-compliance. In Topigs Norsvin (Pty) Ltd v Eskom Holdings SOC Ltd and Others, the Supreme Court of Appeal (“SCA“) considered whether pollution downstream of the Kusile Power Station (“Kusile“) justified court orders compelling Eskom Holdings SOC Ltd (“Eskom“) and the responsible authorities to comply with their environmental duties.

Factual Background

Topigs Norsvin SA (Pty) Ltd (“Topigs“) operates a swine genetics business on a farm adjacent to the Kusile Power Station in Mpumalanga. Eskom is responsible for Kusile’s construction and operation. Kusile’s activities were subject to environmental authorisations, an approved environmental management programme and water-use licences, which imposed requirements governing pollution prevention, water quality monitoring and the management of contaminated water.

Through its representative on Kusile’s Environmental Monitoring Committee, Topigs repeatedly raised concerns about inadequate monitoring and non-compliance. Reports were delayed or deficient and undertakings to address pollution were repeatedly left unfulfilled. Eskom acknowledged that its operations had adversely affected downstream water resources.

Topigs approached the Gauteng Division of the High Court, Pretoria (“High Court”) for declaratory and mandatory relief, together with a structural interdict requiring continuing judicial supervision and reports every 15 days. The High Court dismissed the application, finding that section 28(12) of the National Environmental Management Act 107 of 1998 (“NEMA“) provided an alternative remedy which Topigs had not pursued, and that its reliance on constitutional rights offended the principle of subsidiarity. Topigs then appealed to the SCA.

SCA Appeal

The SCA rejected the contention that Topigs had improperly bypassed environmental legislation by relying on section 24 of the Constitution. The principle of subsidiarity generally requires a litigant to rely on legislation enacted to give effect to a constitutional right. However, Topigs’ case was substantively founded on NEMA, the National Water Act 36 of 1998 (“NWA”) and breaches of Eskom’s authorisations and licences. Furthermore, section 28(12) of NEMA was not an exclusive remedy. Section 32(1) permits an aggrieved person to seek appropriate relief for breaches or threatened breaches of environmental legislation. The failure expressly to invoke section 28(12) did not defeat proceedings firmly grounded in that statutory framework. The SCA found that Topigs had established a clear right, actual or reasonably apprehended harm, and the absence of another satisfactory remedy.

On the merits, Eskom’s persistent non-compliance and the authorities’ inadequate enforcement established breaches of statutory duties and section 24 of the Constitution. The SCA held that the enforcement powers in section 28(4) of NEMA could not be treated as discretionary in the face of established non-compliance. The responsible officials were required to act.

The SCA furthermore rejected the suggestion that this was merely a dispute between neighbours. The protection of water resources and the lawful exercise of public power raised matters of public interest, with pollution capable of affecting communities beyond the adjoining properties.

However, the SCA distinguished between compelling compliance and assuming ongoing supervision of technical remedial work. Updated action plans and steps taken by the departments after litigation commenced indicated a willingness to address their responsibilities. The structural interdict sought by Topigs was therefore inappropriate in the circumstances.

This did not justify refusing all relief. The SCA upheld the appeal in part and granted declarations of unlawfulness and constitutional breach, together with mandatory orders and reporting obligations. Eskom and its current board were directed to submit a detailed action plan under oath within 30 days of the order, identifying the necessary steps, responsible officials and timeframes. They were required to implement the plan without delay and report under oath within 60 days of the order on compliance, explaining any failures and proposed remedial steps.

The Directors-General of the Department of Water and Sanitation and the Department of Forestry, Fisheries and the Environment were ordered to take steps to ensure compliance. Furthermore, within 75 days of the order, they had to file affidavits confirming that they had considered and determined whether enforcement directives, compliance notices or remedial measures should be issued against Eskom under NEMA and the NWA. Topigs was granted leave to respond to Eskom’s reports and approach the court for further compliance relief.

Topigs did not, however, succeed against every respondent. The SCA upheld the misjoinder of three senior Eskom employees. Section 24N(8) of NEMA, which provides for directors’ liability for environmental harm, did not extend to ordinary employees merely because of their seniority. No relief was granted against the cited Ministers or the National Energy Regulator of South Africa, as the necessary basis for relief against them had not been established. Topigs was ordered to pay the employees’ costs. Eskom was ordered to pay Topigs’ High Court costs and the reasonable costs of investigating and preparing the matter, including its expert’s qualifying fees. The successful portion of the appeal also carried a costs award in Topigs’ favour.

Concluding Remarks

The judgment highlights that environmental compliance requires effective implementation and enforcement of the conditions under which a project is authorised. Operators should ensure that remedial plans are implemented and that monitoring complies with the applicable standards. Furthermore, it is apparent that a court may decline continuing judicial supervision while still granting mandatory orders capable of securing compliance. The appropriateness of a structural interdict must therefore be considered separately from the entitlement to other effective relief.

The post Kusile Pollution: SCA Orders Eskom and Regulators to Act appeared first on Werksmans Attorneys.

Artificial intelligence (AI) is ubiquitous, and the gambling and betting sector is no exception. AI is increasingly embedded in gambling and betting products and platforms. The question is not whether to engage with AI, but how to develop, deploy, use and regulate it effectively and responsibly. The policy and regulatory choices made now will influence whether AI is ultimately used to protect consumers or to exploit them.

For gambling operators, AI presents significant opportunities, from processing live information and developing advertising campaigns to creating new features, applications and games. But it also introduces regulatory challenges and risks. AI could support player safety, but it also has the potential to intensify harmful gambling behaviours.

Algorithms allow operators to understand player behaviour and preferences with growing precision. This enables tailored content, services and advertising, while potentially improving the player experience, reducing choice overload, improving stickiness and reducing churn.

AI can also be a powerful responsible-gaming tool, particularly online. Several studies have used player-tracking data and machine-learning models to predict problem gambling using behavioural and monetary indicators. The results have been promising and can inform targeted interventions, including personalised approaches to problem gambling.

Research in France illustrates how this can work. In 2015 and 2016, the French National Gambling Authority, then known as ARJEL, now ANJ, partnered with researchers to develop an online gambling harm-detection system using demographic and behavioural data. The resulting machine-learning models proved highly effective at classifying people by risk level, demonstrating how such research can inform harm-reduction strategies for online gambling.

AI remains an imperfect and developing technology. A key challenge is whether AI systems treat people fairly. AI can perpetuate and exaggerate human biases and inequities, including those based on socio-demographic characteristics. It can hallucinate. Its performance can change over time because of “drift”. It can also simply get things wrong.

AI raises distinct concerns in the gambling context. For example, research has found that large language models repeatedly made irrational, high-risk betting decisions in simulated gambling environments. Because AI systems learn from human-generated data, they can inherit cognitive distortions such as the gambler’s fallacy.

All of this creates a challenge for regulators. South African gambling legislation was drafted for a bricks-and-mortar world. South Africa does have laws protecting consumers in areas including data protection, consumer protection, competition, copyright and cyber law. But it does not yet have AI-specific laws of general application.

Internationally, different approaches are beginning to emerge. The EU AI Act establishes a risk-based classification framework for AI systems. It does not single out gambling, but AI systems used in gambling that fall within high- or limited-risk categories will trigger obligations.

The UK has adopted a cross-sectoral, principles-based approach rather than a single horizontal AI Act. Individual sector regulators, such as the UK Gambling Commission, must apply AI governance principles within their respective domains.

Malta is particularly proactive on this issue. Following consultations earlier this year, on 18 September 2026 the Malta Gaming Authority launched its AI Gaming Charter. The Charter is a sector-specific framework promoting the ethical and responsible use of AI across gaming platforms. It provides a voluntary, principles-based framework to support the responsible, transparent and proportionate use of AI across gaming operations, and practical guidance tailored to the realities of the gaming sector.

In line with the risk-based approach adopted in the EU, the Charter applies a risk-based approach, in which the level of documentation, testing, governance, transparency, human oversight and monitoring applied to an AI system should reflect the nature of the use case and its potential impact on players, player rights, gaming integrity, or regulated outcomes.

The Charter sets out the following principles, which are mapped to the EU AI Act and GDPR, and which are intended to respond directly to the practices, gaps and emerging risks observed across the gaming sector:

• Transparency, accountability and explainability
• Fairness and non-discrimination
• Environmental sustainability
• Data protection, security and governance
• Human oversight and responsibility
• Reliability, safety and robustness
• Compliance, monitoring and AI governance
• Reporting and metrics
• Notification requirements
• Involvement of the Malta Gaming Authority and the Malta Digital Innovation Authority

Existing gambling laws in South Africa and across Africa were not designed for AI. The regulatory gap is real and growing, and the faster AI develops, the harder that gap will be to close.

The post AI and gambling, the regulatory gap is growing appeared first on Werksmans Attorneys.

No one defends the spam caller. According to the National Consumer Commission (“NCC”), South Africans received 17.47 billion spam calls between January and June 2026, a 25.2% increase in the same period during 2025. It therefore comes as no surprise that consumers are frustrated by spam calls from unwanted callers. Nonetheless, it is equally important that a regulatory response must be measured against the harm it targets and the harm it causes. Judged that way, the NCC’s draft Guidelines for Compliance with the Opt-Out Registry Regulations (“the draft Guidelines”) are materially flawed.

Gazetted on 2 October 2026, the draft Guidelines provided interested parties with 15 days to provide their comments for the NCC’s consideration. One could argue that this is insufficient time for interested parties to consider the implication of the guidelines on their businesses and to provide meaningful input. From a simple reading, one finds that these guidelines punish compliant marketers, overlap awkwardly with existing privacy laws, and create new privacy risks of their own with many unanswered questions lingering in the air.

Commercial paralysis

The draft Guidelines require every direct marketer to register before doing any direct marketing, and to renew that registration every year. The net is cast very wide. It covers retailers, insurers, financial institutions, estate agencies, vehicle dealerships and digital marketers, in every industry.

The most disruptive rule concerns consent. The draft Guidelines state that historical or existing consent becomes invalid once a consumer registers a pre-emptive block, and that fresh consent does not override that block unless the consumer first removes it. Practically speaking, this implies that a customer who actively asks a bank or insurer about a new product may not be contacted until that customer has gone back to a state registry and changed their registration. That does not protect consumer choice, but rather overrides it. This could cause a strain on existing customer relationships and how businesses engage with their customers, ultimately creating a barrier to commercial activity between businesses and their customers. Businesses whose growth depends on relationships with existing customers lose their most legitimate marketing channel. Thus, what seems to be a plausible move from the NCC may inadvertently have negative consequences on business, in particular, small and medium sized businesses who rely heavily on direct marketing to gain more customers and generate sales.

A disproportionate burden

The compliance cycle is relentless. Each month, a direct marketer must submit a list of consumers which it intends to contact, pay 12 cents for every name that matches a block, and remove those names. The cleansed list is valid for only 30 days only. For the time being, The NCC has offered free cleansing only from January 2027 to April 2027.

However, for a small business running regular campaigns, this means a recurring cost, an administrative routine to maintain, and the risk of acting on an expired list. The draft Guidelines also “encourages” direct marketers to appoint compliance officers, run training and conduct audits, resulting in businesses accruing more operational expenses. The penalties include administrative fine (either R1 million or 10% of a direct marketer’s annual gross turnover preceding financial year, whichever is great) and, in serious cases, criminal prosecution with imprisonment of up to 12 months.

More uncertainty looms considering that the draft Guidelines set out that they are not binding on the Commission, the National Consumer Tribunal or the Courts, but that anyone applying the Act must take them into account. Businesses are therefore expected to follow rules that the Regulator itself need not necessarily follow nor give much consideration to, creating further legal precariousness and a compliance conundrum. But one has to assume that the Commission would follow the protocol detailed in these Draft Guidelines!

The privacy paradox

The most troubling feature of the draft Guidelines is its approach to data privacy. To protect privacy, it requires direct marketers to upload their customer databases onto a state-run platform every month. POPIA defines an “information matching programme” as the comparison of documents that contain personal information about ten or more data subjects. A national registry that matches millions of records every month fits that description exactly.

Concentrating that much data in one place creates an obvious target for cyberattacks and opens the door for data breaches. It would seem that the NCC has not given much consideration to this issue in light of the fact that the draft Guidelines makes no mention of security safeguards, retention periods, limits on secondary use or what happens after a breach. The Constitutional Court in AmaBhungane Centre for Investigative Journalism NPC and Another v Minister of Justice and Correctional Services and Others struck down legislation in part because it failed to prescribe procedures for how data was to be examined, shared, stored and destroyed. State-held consumer data deserves the same attention. In preparing these draft Guidelines, the NCC somehow overlooked this crucial bit.

Legal and constitutional tensions

The NCC and Information Regulator (“IR“) will seemingly be addressing this issue on a bilateral basis. What does this mean? In essence, the Consumer Protection Act will not operate on its own in curbing unwanted spam calls. Section 69 of POPIA already prohibits electronic direct marketing unless the person has consented or is an existing customer. The IR has said that POPIA compliance remains mandatory whether or not a consumer is on the registry. Direct marketers therefore face two regimes with different consent rules.

POPIA defines consent as a “voluntary, specific and informed expression of will”. The draft Guidelines’ position is that fresh consent cannot override a block contradicts that definition. Section 3(2) of POPIA provides that POPIA applies to the exclusion of other legislation regulating personal information that is materially inconsistent with it. That potentially gives businesses a real argument that the draft Guidelines goes beyond what the law permits.

Importantly, the draft Guidelines present a broader constitutional concern. Section 22 of the Constitution allows the practice of a trade to be regulated, but that regulation must be rational. Courts have set a high bar for striking down economic regulation and will not do so simply because better options exist. Even so, a rule that ignores a consumer’s express, current consent is hard to connect rationally to the stated aim of protecting consumers from unwanted direct marketing.

Lessons from abroad

National do-not-call registries elsewhere, such as those in the United States and the United Kingdom, have shown a consistent pattern. Legitimate businesses comply, while offshore and fraudulent callers ignore the registry altogether. These schemes have generally worked best when they recognise existing-customer relationships and consumer consent, and when enforcement agencies target bad actors rather than imposing a uniform administrative burden on everyone. Our regulators should therefore take note: the scammers generating billions of calls will not register, cleanse lists or pay 12 cents a name nor will the registry inhibit them from calling customers.

A call to action

Time for comment is now. Considering the impact of the Draft Guidelines on marketing activities, requires evaluating the impact thereof on various business units within an organisation. This should be critically evaluated for practicality, failing which business might just be sterilised in terms of marketing activities.

At a minimum, consideration should be given to:

• Recognition of express, current consent and genuine customer relationships, in line with POPIA;
• replace monthly database uploads with a privacy-preserving model in which marketers download suppression lists, supported by published security standards;
• proportionate rules and fees for SMEs; and
• allowing a robust consultation period well beyond 15 days.

Protecting consumers and allowing legitimate commerce are not opposing goals. It would seem as if the current draft Guidelines treats them as such, and both businesses and consumers will pay the price.

What is the saying again about the baby and the bathwater….

Click here to view the draft regulations.

The post Blocking growth? Why the NCC’s Draft Opt-Out Registry Guidelines need revisiting | You were only given 15 days to comment – what are you doing? appeared first on Werksmans Attorneys.

On 7 October 2026, the Constitutional Court (“the Court”) delivered a unanimous judgment addressing a question on which the High Court and the Supreme Court of Appeal had reached opposite conclusions: whether post-commencement creditors are entitled to vote on the adoption of a business rescue plan. The Court held that they are not.

In Salungano Group Ltd v Mashwayi Projects (Pty) Ltd and Others; Ndalamo Coal (Pty) Ltd v Mashwayi Projects (Pty) Ltd and Others [2026] ZACC 38, the Constitutional Court of South Africa upheld the appeals against the decision of the Supreme Court of Appeal and declared that Chapter 6 of the Companies Act 71 of 2008 (“the Act“) does not permit post-commencement creditors to vote on the adoption of a business rescue plan.

The Court emphasised that statutory interpretation cannot be approached by reference to dictionary meanings alone. Instead, the term “creditor” had to be interpreted in light of the text, context and purpose of Chapter 6 of the Act.

While acknowledging that business rescue proceedings do not create a concursus creditorum as in liquidation proceedings, the Court held that the business rescue framework is nonetheless built around a defined body of creditors whose claims existed when business rescue proceedings commenced.

The Court’s decision turned principally on the text and structure of Chapter 6 of the Act.

As a starting point, the Court held that section 145 of the Act only defines three categories of creditors: secured, unsecured and concurrent. This, together with the fact that the term “creditor” was left undefined in Chapter 6, led the Court to conclude that post-commencement creditors are not entitled to vote to adopt a business rescue plan.

The Court also rejected the argument that the reference to “each creditor” in section 145(2) of the Act, which confers the right to vote on a business rescue plan, extends that right to post-commencement creditors. It noted that section 145(1)(c) entitles creditors to participate formally in business rescue proceedings only “to the extent provided for” in Chapter 6, which allows for the possibility that not all creditors may vote. Voting interests are determined by reference to claims admitted or proved in the business rescue proceedings. The mere fact that a person appears in the company’s books as a creditor therefore does not, on its own, confer a right to vote on the plan.

In particular, section 150(2)(a)(ii) of the Act requires a business rescue plan to identify the company’s creditors when the business rescue began, together with details of their ranking in accordance with insolvency law. The Court considered this significant because it establishes a defined creditor body against which voting interests can be determined.

Importantly, the Act does not require post-commencement creditors to be identified in the business rescue plan, nor does it prescribe any mechanism for incorporating their claims into the statutory voting architecture. The Court considered this omission a deliberate choice by the Legislature, and one that leads to the ineluctable conclusion that post-commencement creditors do not get to vote on a business rescue plan.

Who gets a say?

A key aspect of the Court’s reasoning was the distinction between the rights afforded to pre-commencement creditors and the protections afforded to post-commencement financiers.

The Court found that granting post-commencement creditors both preferential repayment rights and voting rights would upset the carefully calibrated balance created by Chapter 6. Pre-commencement creditors are the parties whose claims stand to be compromised by a business rescue plan. By contrast, post-commencement financiers receive statutory protection in respect of the ranking of their claims and are not intended to participate in deciding whether such compromises should occur.

The judgment notes that, because of their enhanced security and preferential status, post-commencement financiers who were also entitled to vote would distort the collective decision-making of pre-commencement creditors in a manner not contemplated by the Act. The Court added that it would be inherently unfair for a creditor whose rights are not curtailed by the plan to vote on a plan that could deprive pre-commencement creditors, particularly unsecured creditors, of a substantial portion of their claims.

The Court emphasised that, even on its reading of the Act, post-commencement financiers remain incentivised to fund business rescue proceedings. Section 135 of the Act grants post-commencement financiers a statutory “super-preference” in respect of repayment. This enhanced protection exists precisely to encourage lenders, suppliers and other stakeholders to continue supporting a financially distressed company during business rescue. The Court viewed this preferential treatment as the legislative bargain struck by Chapter 6: protection through repayment priority rather than governance through voting rights.

Certainty, predictability and the financing trade-off

The Court was also persuaded by practical considerations. It held that allowing post-commencement creditors to vote would create uncertainty because the voting constituency could change as additional debts are incurred during the rescue process. This would make it difficult for business rescue practitioners to determine voting interests and assess whether the statutory voting thresholds had been met.

By limiting voting rights to creditors existing at the commencement of business rescue, the Act promotes certainty, predictability and the efficient administration of rescue proceedings. The Court considered these objectives central to the successful operation of the business rescue regime.

The implications for business rescue stakeholders

This judgment is the final word in the debate relating to the voting rights of post-commencement financiers.

It provides clarity for business rescue practitioners, existing creditors and potential post-commencement financiers, particularly when determining voting interests and during business rescue. There is no doubt that the judgment will shape restructuring and business rescue practice for years to come.

In particular, the judgment confirms that –

• the voting constituency on a business rescue plan is determined by reference to creditors existing when business rescue commenced;
• post-commencement creditors are not entitled to vote on the adoption of a business rescue plan, even though they are creditors of the company;
• while post-commencement creditors are not entitled to vote to adopt a business rescue plan, they are protected by other mechanisms, such as the statutory preference afforded by section 135 of the Act; and
• allowing post-commencement creditors to vote would undermine certainty and predictability in the business rescue process.

The post Voting rights not afforded to post-commencement creditors appeared first on Werksmans Attorneys.

For decades, the detection of cartels has been heavily reliant on a familiar trinity: the leniency applicant who breaks ranks, the disgruntled employee who blows the whistle, and the dawn raid that turns up the incriminating notebook. That model is changing. Competition authorities around the world are now deploying artificial intelligence (AI)  tools to sift through procurement data to detect statistical fingerprints of collusion. This follows a global trend of competition agencies across the world making use of AI tools to detect bid-rigging. It was only a matter of time until South Africa followed in these footsteps.

The South African development

It is reported that the South African Competition Commission is in the process of developing an AI tool to help identify suspected bid-rigging in the local public procurement industry. Commissioner Doris Tshepe told Parliament’s Select Committee on Economic Development and Trade on 30 September 2026 that the tool would assist the Commission in identifying potential cartel conduct in electronic public procurement. The tool is reportedly being developed with the assistance of the Competition and Consumer Commission of Singapore and AI specialists at the University of Pretoria.

This is a logical step considering that countries such as Singapore, Pakistan and Brazil have experienced much success with tools of a same kind, this is according to the Stanford Computational Antitrust’s Fifty Cross-Agency Report which was published in September 2026.

Collusive tendering is one of the forms of horizontal conduct prohibited outright under section 4(1)(b)(iii) of the Competition Act 89 of 1998, and public procurement has long been a priority enforcement area for the Commission. Importantly, the raw material for such a tool already exists: National Treasury publishes procurement data captured through the eTender portal, covering procurement plans, tenders, awards and contracts, in open, reusable formats.

What the rest of the world is doing

The Commission’s initiative places South Africa within a clear and accelerating international trend.

• United Kingdom. The CMA launched an algorithm-based “Screening for Cartels” tool for public procurers in December 2017, designed to flag unusual bidder behaviour and pricing patterns. That downloadable tool was withdrawn in 2020, and the CMA has since built an in-house capability, the Bid Rigging Intelligence Tool (BRIT), which it describes as offering cutting-edge capabilities to analyse public-sector data at scale. In September 2026 the CMA’s Chief Executive estimated that bid-rigging could be costing UK taxpayers between £1 billion and £3.5 billion a year, and the CMA’s Annual Plan 2026 to 2027 commits to using AI and other data science tools to scan bidding data and identify illegal activity at scale. Notably, the deterrent stakes have risen: under the Procurement Act 2023 debarment regime, cartelists face inclusion on a central debarment register and exclusion from public tenders for up to five years.
• Spain. The Spanish competition authority (CNMC) has built its own procurement database and screening tool, BRAVA (Bid Rigging Algorithm for Vigilance in Antitrust), drawing on data that includes losing bids. The CMA reports that the CNMC claims a classification success rate of at least 90%, and that the CNMC has begun developing ATENEA, an agentic reasoning system to analyse and interpret data flagged by BRAVA.
• Brazil and others. According to the CMA, the Brazilian competition authority receives procurement data directly from public bodies, allowing it to apply AI screens to detect cartels. The CMA has also pointed to Portugal and Korea as jurisdictions using data screening tools to find suspicious patterns at scale, and the 2022 compendium referred to similar initiatives in Australia, Canada and the United States.
• European Union. In its 2021 Notice on tools to fight collusion in public procurement (2021/C 91/01), the European Commission encouraged Member States to collect and analyse large volumes of electronic procurement data, “possibly using algorithms, artificial intelligence algorithms or machine learning”, and to give competition authorities access to procurement databases. At the enforcement level, the Commission’s DG Competition has invested in data analysis capability for its ex officio cartel detection work, run machine-learning proofs of concept to support investigations, and leads AI literacy initiatives to ensure that AI is applied safely and effectively in enforcement.

The lesson from abroad: data is everything

The single clearest message from the international experience is that an AI tool is only as good as the data it is fed. The CMA has been candid that BRIT cannot reach its true potential because, unlike in many other countries, losing-bid data is not routinely collected in the UK or made available in a consistent, machine-readable format. Spain’s success, by contrast, rests on a legal requirement to collect specified information about all participating tenderers.

For South Africa, the success of the Commission’s tool will therefore depend less on the sophistication of the algorithm and more on the completeness and quality of the procurement data, including bids by unsuccessful tenderers, pricing detail and bidder ownership information, and on the degree of cooperation between the Commission, National Treasury and organs of state at national, provincial and municipal level.

Questions of fairness and governance

The use of AI by a regulator also raises legitimate questions. An algorithmic flag is a screening signal, not evidence of contravention. Firms may legitimately ask how a tool was trained, what thresholds it applies, and whether its outputs are being treated as a basis for initiating complaints, conducting searches or applying for information. Internationally, regulators are mindful of these concerns: the Commission’s 2021 Notice repeatedly emphasises compliance with data protection law, and the EU AI Act (Regulation (EU) 2024/1689) now imposes a governance framework on certain uses of AI, including by public authorities. In South Africa, the Protection of Personal Information Act 4 of 2013 will be relevant where personal information is processed, and the constitutional requirements of lawful, reasonable and procedurally fair administrative action continue to apply. We would expect the Commission to keep a human investigator firmly in the loop, and to be transparent about the role played by the tool in any proceedings.

What this means for business

Companies that tender for public contracts, and their advisers, should take note:

• Detection risk is rising. Cartel members can no longer assume that conduct will only come to light if a co-conspirator applies for leniency. By way of example, the Swiss Competition Authority has shared that the use of Al tools to detect big rigging has reduced its reliance on whistleblower information to investigate alleged cartel.  Patterns such as bid rotation, cover pricing, consistent price gaps and geographic allocation are exactly what screening tools are designed to detect.
• The leniency calculus changes. As the probability of independent detection increases, the value of being first through the door under the Commission’s Corporate Leniency Policy increases with it. Firms that discover historic misconduct should obtain advice promptly.
• Compliance must be data-aware. Compliance programmes should include periodic reviews of a firm’s own tendering history, careful control of contacts with competitors (including in joint ventures, subcontracting and consortium bids), and documentation of independent bid determination.
• Consequences go beyond fines. Beyond administrative penalties of up to 10% of turnover, bid-rigging can expose firms to procurement exclusions, civil damages claims and, for directors and managers, criminal liability under section 73A of the Competition Act.

Conclusion

The Commission’s AI initiative, developed with international and academic partners, reflects a global shift from reactive to proactive cartel enforcement. If properly resourced, supported by good-quality procurement data and applied with appropriate safeguards, it has the potential to change the risk profile of public procurement in South Africa significantly. For firms that participate in public tenders, the time to review competition compliance is now, before the algorithm does it for them. In our view, this signals the beginning of new era for competition compliance in South Africa, mainly that Al tools will have a meaningful impact on the enforcement of competition law against anti-competitive practices. What lies ahead however…remains to be seen.

The post The algorithm will see you now: Competition Regulators turn to AI to catch bid-riggers appeared first on Werksmans Attorneys.