---
title: "South Africa is the land of AI contradiction when it comes to jobs"
description: "The local AI conversation has moved from job displacement panic towards autonomous AI agents, unmanaged ‘Shadow AI’, and emerging cybersecurity vulnerabilities that threaten corporate governance, data sovereignty and Popia compliance."
type: "NewsArticle"
publisher: "Daily Maverick"
site: "https://www.dailymaverick.co.za"
section: "THE DIGITAL FRONTIER"
author: "Lindsey Schutters"
author_url: "https://www.dailymaverick.co.za/author/lindsey-schutters-2/"
canonical_url: "https://www.dailymaverick.co.za/article/2026-10-06-ai/"
published: "2026-10-06T22:41:28"
lang: "en-ZA"
word_count: 1035
---

# South Africa is the land of AI contradiction when it comes to jobs

> The local AI conversation has moved from job displacement panic towards autonomous AI agents, unmanaged ‘Shadow AI’, and emerging cybersecurity vulnerabilities that threaten corporate governance, data sovereignty and Popia compliance.

By Lindsey Schutters · Published 7 October 2026, 00:41 SAST

## Key points
- South Africa faces a generational AI paradox: near-record unemployment alongside the continent’s fastest adoption of generative AI.
- While 23.1% of working-age South Africans use AI, only 0.6% report formal AI skills — leaving young people, especially women, exposed as entry-level jobs vanish.
- The next threat is already inside companies: unsanctioned ‘Shadow AI’ and autonomous agents can access sensitive data, breach Popia compliance and trigger costly cyberattacks.
- Experts call for urgent investment in education, reskilling and human oversight — including turning indigenous-language data projects into jobs and a sovereign national asset.

## Content

There’s [something happening here, and what it is may not be exactly clear](https://www.dailymaverick.co.za/video/2026-09-30-watch-debating-ais-existential-risk/). You know the stat that gets fired off often well: out of 103 economies tracked by the International Monetary Fund (IMF), South Africa ranks second-highest in unemployment globally, behind only war-torn Sudan.

Further to that – and moving beyond the devastating youth unemployment statistics – 3.9 million young people aged 15-24 (37.6%) are classified as NEET (Not in Employment, Education or Training), with a youth labour absorption rate of just 10.1%.

[ARTIFICIAL INTELLIGENCE

South Africa on the road to building a sustainable AI tech ecosystem

April 1, 2025 ![Image](https://cdn.dailymaverick.co.za/i/P4gNmeyVzCJ3nfjn5cqKQMqJh7E=/450x0/smart/file/dailymaverick/wp-content/uploads/2025/03/GettyImages-2121060609.jpg)](https://www.dailymaverick.co.za/article/2025-04-01-sa-must-learn-from-chinas-digital-white-elephants-and-build-a-sustainable-ai-tech-ecosystem/)

But despite these systemic labour market challenges, South Africans are adopting generative AI rapidly. According to the Microsoft Global AI Diffusion monitor, South Africa leads the continent with 23.1% of the working-age population actively using AI in daily activities.

However, that does not translate into more opportunity for work. Formal AI skills penetration remains severely lagged: the Stanford University HAI AI Index reveals that self-reported AI skill penetration among South African workers stands at just 0.6%, compared with 2.8% in India, 2.2% in the United States, and 1.9% in Germany.

[YouTube video: "YouTube video"](https://www.youtube.com/watch?v=z-7Tq6gWWVc)

### **What’s that sound?**

“The stakes are generational. No economy can easily absorb even a 5%-7% uptick in unemployment over a few months; that is civilisational-stress territory if you are not prepared,” Professor Benjamin Rosman, chair of South Africa’s National AI Advisory Panel & TIME100 AI honouree, said in a new report on AI literacy ([Walking the tightrope: South Africa’s AI future](https://www.thecollectivex.org/wp-content/uploads/2026/10/Artificial-Iintelligence-jobs-and-skills-in-South-Africa-FINAL-RGB-FOR-WEBSITE.pdf)) by Collective X.

Modern job threats change too fast for legislators to clearly define or solve.

“AI is neither inherently job-destroying nor job-creating,” Dr Rodney Manyike of the Human Resource Development Council explains about the policy dilemma.

“Its impact will depend largely on the policy choices South Africa makes today. The deciding factor is not the technology itself, but the speed and seriousness of the country’s investment in education, reskilling, workforce transition programmes, and the policy choices and positioning that will accompany AI adoption.”

Data reveals that roles most vulnerable to immediate automation are entry-level positions where young people and young women predominantly begin their careers. If these pathways disappear without replacement, social instability and economic exclusion will deepen.

It starts from the top. Major technology hyperscalers are spending upwards of $1.4-trillion annually on capital expenditures, creating an incentive to replace high worker wage bills with algorithms to generate returns on these massive investments.

[For context

Hallucination-free (hopefully) — new draft AI policy will be ready by March, says Malatsi

September 21, 2026 ![Image](https://cdn.dailymaverick.co.za/i/6_zFB3r3AEg2W4Nbq5fY5E9j-o8=/450x0/smart/file/attachments/orphans/BMwisdomofSolly1_345574.jpg)](https://www.dailymaverick.co.za/article/2026-09-21-hallucination-free-hopefully-new-draft-ai-policy-will-be-ready-by-march-says-malatsi/)

### **Everybody look what’s going down**

To counter pure displacement, South African experts advocate labour-augmenting, human-in-the-loop framework deployment.

“From the day the algorithm is being written until it is deployed and used by the end user and continuously being used, you need human intervention,” is the view of Monisha Prem, a lawyer, AI researcher and educator who works within AI governance and skills development through her legal firm in Johannesburg. “The more checkpoints you have, the closer you are to controlling it.”

Researchers are proposing a hyperlocal temporary fix. Because foreign AI models show a 31% performance drop in African languages compared with English due to massive data scarcity, monetising local language data could be an answer.

On a token use or just processing cost curve, official languages like isiZulu, isiXhosa, and Sesotho have only 230MB to 290MB of online text each, compared with 9TB for English; a 10,000-to-1 ratio. That’s according to IrokoBench, the benchmark built by South Africa-based AI collective Masakhane, but even if the numbers are skewed, it is still a significant statistic.

![AI](https://cdn.dailymaverick.co.za/i/azfhAvhFWB0_Uz2vvxf85q6QmL0=/200x100/smart/filters:strip_exif\(\)/file/attachments/2998/Screenshot2026-10-06at193435_781484.png)

*Source: [Walking the tightrope: South Africa’s AI future](https://www.thecollectivex.org/wp-content/uploads/2026/10/Artificial-Iintelligence-jobs-and-skills-in-South-Africa-FINAL-RGB-FOR-WEBSITE.pdf) by Collective X*

Establishing national, sovereign programs to transcribe and annotate indigenous language data creates thousands of dignified, paid opportunities for rural and township language speakers while building a permanent national asset.

But as South African businesses move beyond basic text-prompt chatbots (read: ChatGPT, Claude) into operational AI, the enterprise conversation has turned sharply towards autonomous AI agents and cybersecurity risks.

“The number one danger of AI inside organisations isn’t the technology itself – it’s AI that nobody is watching,” explains Chris Badenhorst, head of Azure core, data and AI services at Braintree.

“Shadow AI” happens when employees, trying to be more productive, paste sensitive company and customer data into unsanctioned tools, or spin up AI agents that act on the data with no oversight.

“The risk is growing fast as businesses move from chatbots to autonomous agents that can read email, access files and take actions on someone’s behalf. Each of those agents is effectively a new digital worker with access to your information, and most organisations have no inventory of how many they have, what those agents can reach or what they’re doing.”

### **Step out of line**

Cybersecurity telemetry from Eset indicates that between March and May 2026, the number of unique AI skills scanned surged from 60,000 to nearly 900,000, with more than 25,000 flagged as suspicious and more than 3,000 identified as explicitly malicious.

Across 13 frontier models evaluated in security stress tests, every single model was successfully hijacked at least once across 250,000 attack attempts.

You have obviously heard about the Hugging Face hack and the Australian Department of Health data breach. Allan Juma, lead cybersecurity engineer at Eset warns that autonomous agents operating with standing permissions create severe operational risks.

For example, if an internal insurance model autonomously alters customer policy parameters following an indirect prompt injection or logic flaw, it creates immediate financial, operational and regulatory crises.

![AI](https://cdn.dailymaverick.co.za/i/-yg9mwC-4jD5kcsvvGjSODZViLc=/200x100/smart/filters:strip_exif\(\)/file/attachments/2998/Screenshot2026-10-06at193326_510719.png)

*Source: [Walking the tightrope: South Africa’s AI future](https://www.thecollectivex.org/wp-content/uploads/2026/10/Artificial-Iintelligence-jobs-and-skills-in-South-Africa-FINAL-RGB-FOR-WEBSITE.pdf) by Collective X*

The answers? There are no simple ones. In Mzansi, all fingers point to educator capacity, not learner appetite, as the bottleneck to building national AI capacity.

Enterprises must also start treating every AI agent as a digital employee, and assign them distinct digital identities that can help limit data permissions strictly to required tasks, and enforce continuous behavioural monitoring.

Our future workforce is already existing alongside AI agents and solutions; we need to start redesigning our workplaces for this reality.

There’s a digital man with a gun over there, and our regulators and policymakers have got to beware. **DM**