Is South Africa’s AI Teaching Robot a Scam?
Government celebrated the launch of an AI teaching robot factory in Durban on 13 August 2026. The factory does not exist yet. The robots are built in India. The entry price for a single unit is R250,000, with components that in my assessment cost under R25,000. And the technology was passed over in the country it came from before South Africa imported it.
On 13 August 2026, the Department of Science, Technology and Innovation helped launch what its own press release described as a manufacturing plant and showroom at Dube TradePort in Durban. The product is IRIS, a humanoid robot tutor from a company called BSG and Technologies, founded by former KwaZulu-Natal teacher Thando Gumede. Deputy Minister Nomalungelo Gina said the initiative has the potential to create 20,000 jobs over the next 5 years. The company’s stated goal is IRIS in all 23,000 schools in South Africa.
Those are the claims. What follows is what the record actually shows.
The factory is a plan, not a building
Local reporting from the launch itself says IRIS is currently manufactured in India, and that BSG and Technologies has ambitions to manufacture in South Africa one day, with plans to establish a facility at Dube TradePort. The department’s own press release calls IRIS locally developed. The machines in South African classrooms come off a production line in India.
The same reporting says the company employs 15 people. A promise of 20,000 jobs rests on a company of 15 people, with no factory, selling a robot built in another country. Government celebrated the plan, not the plant.
This is not an isolated observation. Deputy Minister Gina said from the podium at the same event: “I am also encouraged by the prospect that this humanoid will no longer be built entirely outside South Africa.” That is the government itself confirming, in the speech used to announce the factory, that the robot has been built entirely outside South Africa.
She also said: “If the intellectual property, technical skills, maintenance capability and production capacity are increasingly located in South Africa, then investment in IRIS becomes more than the purchase of a device.” The word “if” is carrying a great deal of weight in a sentence attached to a factory launch.
What IRIS actually is
DSTI’s own media statement from the original 2025 launch described IRIS as a system equipped with software and a keyboard, using voice rather than text, functioning similarly to any AI tool using deep machine learning. By the department’s own account, the core of IRIS is a voice-driven AI assistant. It answers questions, assists with homework and revisions, is claimed to mark work and analyse learner performance. Those are significant claims. No independent auditor or IT professional has assessed them in a controlled classroom setting. Every demonstration on record has taken place at a media showcase with the developer present.
BSG and Technologies lists a co-founder on its own website: Makerlabs, an education technology company based in India. Makerlabs launched a robot also called IRIS at a school in Kerala in March 2024, more than a year before the South African version appeared. Same name, same concept. India called it the world’s first AI teacher robot. South Africa called its version the country’s first. Both claims cannot be right, and the co-founder relationship between the 2 companies suggests the South African version is not a separate invention.
What India did next
The Kerala robot went into a single private school through a government grant for school technology labs. That school was selected because of its profile at science and youth festivals, not through any merit-based educational assessment. A similar unit later appeared at a school in Assam under the same programme.
No independent study of what either robot did for children’s results exists in the public record. Every article about IRIS in India is a launch-day story. After more than 2 years, there is no published evidence that the robot added measurable educational value at either placement.
Kerala’s own state education department then voted with its budget. In February 2025 it announced it would build its own AI system for schools as software, and had already distributed 29,000 basic robotics kits with no humanoid body at all. The state where IRIS originated looked at the robot and chose a simpler, cheaper, non-humanoid alternative for its own rollout. South Africa is importing the version its home market declined to scale.
Where the robots are going
5 schools have received IRIS units through what BSG calls the National IRIS Lab School Campaign. The schools are Amangwe High School, Mgidla Secondary School, Ekuthuthukeni Secondary School, George Caltex Primary School and Intakemazolo Combined School, all in KwaZulu-Natal.
South Africa ranks schools in quintiles from 1 to 5, with quintile 1 being the poorest and quintile 5 the wealthiest. Amangwe High School and Intakemazolo Combined School are quintile 3. Mgidla Secondary is quintile 2. Ekuthuthukeni Secondary and George Caltex Primary are both quintile 1.
Government’s own connectivity policy says the poorest schools come first because wealthier schools can raise their own technology funding. On paper, the pilot targeting is broadly in the right direction. But the method of selection raises its own questions.
BSG’s own presentation materials show the National IRIS Lab School Campaign allocates units through a community mobilisation and voting process, where schools campaign for votes among learners, parents, alumni, traditional leaders and local communities. Schools do not receive an IRIS because they have the greatest educational need. They receive one because they mobilised the most votes. That rewards schools with active alumni networks and social media reach, which is not the same thing as greatest need.
The pilot proves nothing
Gumede has said IRIS needs WiFi and a stable electricity supply, with a back-up battery for gaps. Where schools have no internet, her company says it has partnered with businesses to provide connectivity.
A pilot where the developer selects the schools and private companies patch the infrastructure gaps is a demonstration, not a stress test. It tells you nothing about school number 6,000, where there is no launch event, no developer on site, and no sponsor paying the data bill. The question that matters is not whether IRIS works under hand-picked, closely supported conditions. It is whether it works when that support is removed.
The infrastructure problem
The DA revealed through a parliamentary question that Gauteng learners lost 8,925 school days in 2026 because schools had their electricity cut off. Gauteng is the wealthiest province in South Africa.
One of those schools is Boitumelong Secondary in Tembisa, launched by Premier Panyaza Lesufi as a fully digital school with smart boards, free WiFi and tablets. This year Boitumelong sat in the dark over a R1 million electricity debt to the Ekurhuleni metro. Pupils went back to textbooks and whiteboards, waiting for solar power to switch on around 11am.
If a flagship digital school in Gauteng cannot keep its smart boards running because of a municipal billing failure, the infrastructure risk for a WiFi-dependent robot in rural KwaZulu-Natal is not theoretical. It is the documented, repeated experience of every technology rollout government has attempted in under-resourced schools.
Internet connectivity compounds this. Rural school connectivity is close to nonexistent at most sites, and no nationally funded solution to that gap is in place. Starlink has effectively been made unavailable in this country, so the burden falls on other providers who have not stepped in at scale either.
The theft question nobody answered
Parliament’s committee on school connectivity asked the Department of Basic Education a direct question: of the 545,938 learner devices and 30,818 teacher devices procured for schools, how many were lost or stolen? No number was provided.
What the department has said in written replies to Parliament is that the rollout of devices and internet to schools is held back by a lack of funding, by theft, by vandalism and by maintenance costs. The department has also confirmed that the entire programme is an unfunded mandate. There is no national budget allocated for it. Provinces carry it from existing budgets.
Government has not funded the laptops and internet it already promised schools. It has now celebrated a robot that needs both to operate.
The price nobody mentioned at the launch
BSG and Technologies has published its pricing. The entry package, called Early Adopter, costs R250,000. It includes 1 IRIS robot, core curriculum software, on-site installation and staff training, 12 months of hardware and software support, and access to all South African languages.
The middle package, called Connected Classroom and marked as the most popular, costs R325,000. It adds 16 tablets with the IRIS Student App and BSG Learning Management System pre-installed, full system integration, on-site installation and staff training, and 24 months of premium support.
The top package, Smart Classroom Integration, costs R430,000. It adds a single 86-inch interactive smartboard and 36 months of support.
All 3 packages are listed as preorders. No school has paid yet.
Now apply those numbers to the stated goal of 23,000 schools.
At R250,000 per school, the full rollout costs R5.75 billion. At R325,000 per school, it costs R7.475 billion. At R430,000 per school, it costs R9.89 billion. The entire national Basic Education infrastructure budget for 2025/26 was R22.6 billion. This robot rollout at the entry price alone would consume more than 25% of that budget. At the middle package it would consume 33%.
There is a further comparison worth making. A starting teacher in South Africa earns roughly R200,000 to R220,000 per year. R250,000 buys 1 robot with 12 months of support, serving a school of over 1,000 learners. The same money pays for more than a year of a qualified human teacher in a classroom.
And then there is the hardware itself.
I have watched the demonstrations. I have looked at the components visible in the photos and videos from the launch events and from comparable systems. In my assessment, the hardware inside an IRIS unit costs under R25,000. That is my observation and I am willing to stand behind it. If I am wrong, BSG and Technologies can publish a full component breakdown and cost justification, and I will cover it.
Which means the rest of that R250,000 entry price is not technology. It is the humanoid shell, the branding, the launch events, and the cost of selling something that looks like a robot rather than what it actually is, which by the government’s own description is a voice assistant with a keyboard.
The argument for parents
If the goal is personalised, one-on-one AI tutoring, a humanoid robot is the wrong shape for the job. Ekuthuthukeni Secondary has over 1,000 learners and 1 IRIS unit. Dozens of children waiting their turn at a single machine is a demonstration, not a tutoring programme.
The same AI tutoring function delivered as software on a standard laptop or desktop changes that ratio completely. Government has already procured over 545,000 learner devices. It already operates a system called the C3, a small local server that sits inside a school and functions without internet access. Software on that existing infrastructure gives every child an individual session instead of a queue. It uses a fraction of the electricity. It has no motors, joints or wheels to break. And a laptop is a far smaller theft target than a humanoid machine.
BSG and Technologies already lists computers, laptops and digital learning resources among its own services. The cheaper delivery method is already on the company’s own service list. The robot is the headline product. The laptop is the footnote.
A screen with a voice delivers the same educational function as a mannequin with a screen for a face. The humanoid body adds fragility, cost, power consumption and theft risk. The humanoid body is a marketing decision. And at R250,000 per school, somebody has to pay for it.
What I am asking
I am not dismissing the intention behind this project. The teacher vacancy crisis is real. 27,092 posts are vacant as of 30 June 2026. The Eastern Cape alone carries 5,730 of them. If technology can genuinely assist where teachers are absent, that is worth pursuing.
But the pricing is now on the table. R250,000 for a robot whose components I assess at under R25,000, targeted at quintile 1 and 2 schools, with no national budget to pay for it, no proven results from its country of origin, and a pilot that has been supported by the developer and private sponsors rather than tested in the conditions it would actually face at scale.
Before this project goes further, BSG and Technologies and the Department of Science, Technology and Innovation owe the public a full accounting. What is inside the unit and what does each component cost? What independent evidence exists that IRIS improves learning outcomes? Who is paying for the pilot schools, and who will pay for the rest? And why does a voice assistant with a keyboard need to look human and cost R250,000 to do it?
The current proof of what this system can do is limited to media showcases. A full demonstration, assessed by an independent auditor and IT professional in a real classroom without the developer present, is not an unreasonable ask for a project claiming 23,000 schools and 20,000 jobs on the back of a showroom and a plan.
