AI Will Not Stop the Cable Thieves — and South Africa's Rail Revival Depends on Understanding That

The technology being deployed on the newly privatised rail network can optimise every system it touches. The organised criminal networks stripping it bare are outside its reach.

Sign up for my Substack daily AI newsletter here.

See my AI Training course portfolio for corporate Business Leaders here.

Article link: https://open.substack.com/pub/johanosteyn/p/ai-will-not-stop-the-cable-thieves?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

In May 2026, South Africa did something it had not done in more than a century. The Transnet Rail Infrastructure Manager finalised access agreements with 11 private Train Operating Companies, allocating slots across five strategic freight corridors and immediately adding 24 million tonnes of freight capacity to a network that had been haemorrhaging volume for a decade. The allocations span coal from Mpumalanga to Richards Bay, iron ore from the Northern Cape, containers, fuel, and general freight. The potential exists to scale to 52 million additional tonnes over five years, supporting the national objective of increasing rail volumes from approximately 180 million tonnes to 250 million tonnes by 2030. TRIM Chief Executive Moshe Motlohi described it precisely: the creation of a functional and competitive rail marketplace, moving from policy design to practical implementation.

Simultaneously, Transnet Freight Rail has issued a tender for AI-Assisted Operational Intelligence — a system capable of ingesting data from IoT sensors, maintenance records, and operational databases, applying machine learning for anomaly detection and predictive maintenance, and providing real-time decision support for network management. This is the right technology for the right problem. The question that deserves more attention than it is receiving is whether the problem AI is being asked to solve is the most important one on the network.

CONTEXT AND BACKGROUND

Transnet’s freight volumes peaked at 226.3 million tonnes in 2017. By 2022 they had fallen by 34% to approximately 149.5 million tonnes — a level not seen since World War II. The causes are well documented: underinvestment, mismanagement, equipment shortages, and a decade of state capture that hollowed out the institution’s capacity to maintain, renew, and operate its infrastructure. But the immediate operational cause, the one that stops trains in real time and makes the network unreliable for the exporters who depend on it, is physical. Cable theft and infrastructure vandalism have caused damage and disruption at a scale that defies description in conventional infrastructure management terms.

Since 2019, more than 4,600 kilometres of copper cable have been stolen from Transnet’s network. In the 2021 to 2022 financial year alone, Transnet lost R1.8 billion to cable theft, losing approximately 100 kilometres of cable per month. By 2022, the Transnet annual report documented a 1,096% increase in cable theft over five years, with more than 1,500 kilometres stolen and a net financial impact of R4.1 billion. Corruption Watch estimated the total economic impact of copper theft on South Africa’s rail and electricity networks combined at more than R45 billion in the year to March 2022. The GAIN Group calculated that Transnet’s operational failures were costing the South African economy approximately R1 billion a day in 2023 — nearly 5 per cent of annual GDP.

The consequences are not abstract. When copper cable is cut from a rail line, the train stops. It stops until a maintenance crew can be dispatched to the location, the damage assessed, and the cable replaced. On the Durban to Johannesburg corridor — designed to transport 70 trains daily — this has at times reduced throughput to fewer than ten trains per day. The iron ore export line from the Northern Cape to Saldanha Bay, historically the network’s best-performing bulk line, has faced repeated service collapses directly attributable to cable theft. The Africa Report has documented that these thefts are not opportunistic. They are organised. Gavin Kelly, CEO of the Road Freight Association, is direct: an organised syndicate is usually behind copper theft on rail lines. At the scale and frequency Transnet has experienced, this is not a maintenance problem. It is a security problem.

INSIGHT AND ANALYSIS

Transnet’s AI tender is the correct strategic response to a well-governed, operationally stable rail network seeking to move from reactive maintenance to predictive intelligence. The tender’s specifications are sound — machine learning for anomaly detection, predictive models for rolling stock and infrastructure, self-learning capabilities that improve over time, integration with IoT sensors and legacy operational databases.

The global evidence for AI in freight rail is compelling — but it was generated under conditions that differ materially from those facing South African corridors. Understanding that gap is the most important strategic insight available to any board or executive evaluating this investment.

Where Predictive AI Succeeds: Stable Global Environments

Rio Tinto’s AutoHaul system operates 200 autonomous locomotives across 1,700 kilometres of track in the Pilbara region of Australia, eliminating the need for drivers and generating billions in additional free cash flow. European and North American rail operators have documented significant efficiency gains from predictive maintenance systems that replace fixed-interval maintenance schedules with condition-based interventions. In these environments, AI thrives because the foundational conditions are met: the physical infrastructure is present and intact, the primary operational risks are equipment wear and weather, the network is not being stripped of its signalling infrastructure between train movements, and data pipelines are continuous, consistent, and uninterrupted by criminal interference. AI can predict what it can measure. In stable environments, it measures everything that matters.

Where Predictive AI Reaches Its Limits: Volatile Trench Realities

South Africa’s rail corridors present a fundamentally different operational context. Cable theft is not a maintenance event that AI can anticipate. It is a deliberate, organised act by criminal syndicates that has removed more than 4,600 kilometres of copper from the network since 2019. When copper cable is cut, trains stop — not because of a bearing failure the algorithm could have flagged three weeks earlier, but because the physical medium that carries the signalling data no longer exists. AI can tell you that a bearing is approaching failure. It cannot tell you that a criminal syndicate will cut the signalling cable on a specific section of track at 2am on a specific night. AI can optimise train scheduling around known network conditions. It cannot schedule around the discovery that 50 kilometres of copper cable has been stolen from a corridor overnight. AI can reduce the cost and frequency of planned maintenance interventions. It cannot replace the physical cable, the relay equipment, or the substation components that are removed from the network by organised theft at rates that have at times reached 100 kilometres per month. A predictive maintenance system is valuable when the asset it is monitoring exists. When that asset has been removed by a criminal network, the algorithm has nothing to predict.

This is not an argument against Transnet’s AI strategy. It is an argument for sequencing and for honest assessment of what AI can and cannot contribute to this specific revival effort.

IMPLICATIONS

To be precise about the reform’s legal architecture: South Africa has not sold off its rail assets. Under the open-access, multi-operator framework established by TRIM, the state retains full ownership of the rail infrastructure. Private operators purchase operational slots and run their own locomotives and wagons on the network, paying access fees for allocated capacity. The distinction matters because the obligations that flow from infrastructure ownership — including the obligation to maintain and secure that infrastructure against organised criminal damage — remain with the state, not with the private operators now competing for freight business on the network.

The eleven Train Operating Companies that have been allocated slots bring capital, technology, and competitive incentive. They also bring a legitimate expectation that the infrastructure they are paying to operate on will be available, functional, and secure. That expectation is reasonable. It is also not guaranteed by the reforms that have been implemented, because the open-access framework addresses competition and operational efficiency but does not directly address the physical security of the corridors those operators are being invited to compete on.

The most important governance question facing South Africa’s rail revival is therefore not which AI vendor wins Transnet’s operational intelligence tender. It is whether the physical security of the rail network — the cables, the signalling equipment, the relay infrastructure, the substations — is being addressed at the speed and scale that private operators, the African Development Bank’s 1 billion dollar loan facility, and the national freight volume targets require. Cable theft reduction requires law enforcement capacity, prosecutorial will, scrap metal dealer regulation, community engagement in the areas where theft is most concentrated, and the deployment of surveillance technology including the drone monitoring that has already shown early results on the ore line. Some of these are Transnet’s to address. Others belong to the South African Police Service, the National Prosecuting Authority, and the municipalities through whose jurisdictions the rail corridors pass.

For boards and executives whose organisations depend on South Africa’s freight rail network — mining companies, agricultural exporters, manufacturers, logistics providers — the open-access reform and AI investment represent a genuinely consequential shift in the right direction. The strategic question is whether the productivity dividend those investments are designed to deliver will be realised on the timelines being projected, given the continued presence of the security threat that drove volume from 226 million tonnes to 149 million tonnes in the first place. AI and open-access competition together can produce a world-class freight rail network. They cannot do it on a network that is being physically dismantled faster than it can be maintained.

CLOSING TAKEAWAY

South Africa’s decision to open its rail network to multi-operator competition and to deploy AI-driven operational intelligence is the right strategy for the network this country needs. The honest assessment of its likely success requires acknowledging the condition it is arriving at. The network that private operators are being invited to compete on, and that AI systems are being asked to optimise, is one that lost a third of its freight volume over a decade to theft, vandalism, and the organised criminal networks that profited from both.

Technology can do extraordinary things for a rail network. It can predict failures before they occur, optimise asset utilisation, and reduce the cost of maintenance at every point in the system. What it cannot do is substitute for the physical security of the infrastructure it is meant to manage. South Africa’s rail revival depends on understanding that distinction — and on ensuring that the investments being made in AI and open-access competition are matched by investments in the one condition that makes both worthwhile.

Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net