South Africans’ street smarts as a result of the high levels of crime, as well as a low per capita gross domestic product (GDP), are proving to provide a level of protection against falling victim to AI-enabled fraud.
This is the opinion of Alistair Pugin, founder of @Celyntis, a cloud technology enabler, who says South Africans aren’t immune to AI-enabled fraud – they get caught too – but a culture shaped by high crime may make them somewhat more sceptical targets than people in wealthier markets.
“We worry about everything; we look over our shoulders when we’re driving down the road. Our level of gullibility is not as high as the rest of the world,” says Pugin. “Our level of crime is a blessing in disguise.”
Pugin was commenting on the fact that SA accounts for less than 1% of all cases recorded on the Artificial Intelligence (AI) Incident Database (AIID) between March 2023 and April 2026.
These incidents are diverse and point to an evolving use of the technology for fraud, manipulation, impersonation and the manufacture of false information.
However, as the landscape evolves, with increasingly sophisticated scams, more people coming online through the expansion of broadband, cheaper data and growth in GDP, this situation could change, cautions Pugin.
He believes this could expose more South Africans to AI-enabled threats, potentially eroding the relative resilience that currently comes from limited access and greater caution.
“We’ve never seen a rate of change in any industry in the history of humankind that we’ve seen with AI,” says Pugin.
He points to Microsoft CEO Satya Nadella’s recent argument that Moore’s Law is becoming outdated as the pace of computing advances accelerates, with capabilities now doubling around every six months rather than the traditional two-year cycle.
By the numbers
Of the 1 618 cases collected by the database, ITWeb’s AI-assisted research shows SA accounts for 12, although one incident is recorded twice. Although the AIID database isn’t a final authority on the volume of AI-related crime, the Massachusetts Institute of Technology (MIT) records a similar number at 1 400.
SA’s 12 incidents extend beyond hacking, showing a progression from AI unintentionally producing false information that others relied on, to AI being deliberately used to fabricate falsehoods, impersonate people, facilitate fraud and manipulate real-world events and systems. MIT’s AI Incident Tracker – which pulls raw data from several databases – shows a similar trend.
AIID records incidents where an AI system played an important causal role in real-world harm – a “but-for” cause, whether the harm is financial, physical, psychological, reputational, political, social or related to civil liberties – regardless of whether that harm was accidental or deliberate.
As a result, the SA list contains items that are all legitimately AI incidents, although they involve completely different mechanisms. Also included for the sake of completeness are two that impacted people outside of SA but originated here.
Moreover, they show AI moving across four distinct risk surfaces: criminal enablement (scams, malicious sites and impersonation), information integrity (deepfakes, fabricated images and misinformation), institutional trust (fake legal citations and government references), and critical services (biometric systems used to determine access to grants).
The South African connection also isn’t one-directional, as SA appears variously as victim, target, institution, actor and alleged source of AI-enabled activity. AIID lists the below incidents in its database as being linked to SA:
Grok and SA narratives
In May 2025, xAI’s Grok began inserting unsolicited references to “white genocide” in South Africa into a wide range of otherwise unrelated conversations on X.
The model wasn’t just responding inaccurately to questions about South Africa; it was introducing inflammatory, racially-charged material into conversations where it was not relevant, affecting X users, public discourse and Black South Africans.
AI-generated clickbait
An AI-enabled scam targeting teenagers in Rockingham County, North Carolina, used fake local headlines and a student’s likeness to lure users to malicious websites.
Investigators said nearly 100 sites were linked to a centralised operation in South Africa, with the sites deploying harmful advertising or malware once they identified visitors as students or local residents.
Fabricated case law in SA
One of the earliest incidents in the dataset dates to March 2023, when lawyers representing a plaintiff in a Johannesburg Court defamation matter relied on legal judgements generated by ChatGPT.
The cases and citations were fictitious, forcing the court to delay proceedings while the references were checked; the legal team subsequently faced a punitive costs order and criticism for failing to verify AI-generated research.
The Mavundla case
In Mavundla v MEC, lawyers representing South African politician Philani Godfrey Mavundla submitted citations that appeared to have been generated by AI, only for the Pietermaritzburg High Court to find that many of the cases did not exist.
The matter ultimately resulted in the firm being referred to the Legal Practice Council.
Legal genius
In June 2025, another South African legal team acknowledged that non-existent case citations submitted in an urgent licensing dispute had been produced by Legal Genius AI.
The Johannesburg High Court found that the fictitious references undermined the credibility of the submissions and referred the lawyers’ conduct to the Legal Practice Council for investigation.
Trevor Manuel deepfake
AI was put to a more overtly criminal use in an investment scam involving purported deepfake videos of former finance minister and then Old Mutual chairman Trevor Manuel.
The videos used his likeness and cloned voice to promote fake investment schemes on social media, prompting Manuel to publicly deny any involvement.
Thai election disinformation
Ahead of Thailand’s February 2026 election, an apparently AI-generated image circulated purporting to show Thai prime minister Anutin Charnvirakul dining with South African businessman Benjamin Mauerberger – implying a non-existent relationship.
Google’s SynthID confirmed it was highly likely it was made with Google AI tools.
AI hallucinations in SA
South Africa’s draft National AI Policy was found to contain fictitious references believed to be AI hallucinations, while the Department of Home Affairs later withdrew an apparently AI-generated list of fictitious references appended to its Cabinet-approved Revised White Paper on Citizenship, Immigration and Refugee Protection.
Home Affairs suspended two senior officials and commissioned an independent review of policy documents produced since November 2022.
Iranian prisoners
AI-enhanced images of eight Iranian women political prisoners were circulated alongside an unverified claim that they faced imminent execution; the Iranian Embassy in South Africa was among the actors listed by AIID in connection with the incident.
Fake school fire
In January 2026, a purportedly AI-generated image showing Orlando West High School in Soweto engulfed in flames was shared with emergency services.
Officials said the false report caused public alarm and diverted emergency resources.
SASSA facial verification failures
The South African Social Security Agency’s eLife Certification portal experienced repeated facial-recognition and OTP failures, leaving beneficiaries unable to complete mandatory verification and potentially putting grants at risk. A later parliamentary report confirmed 67 868 grant suspensions during one quarter.
Vanderbijlpark crash
After a fatal school-transport crash near Vanderbijlpark in January 2026, AI-generated memorial posters began circulating on Facebook, X and TikTok showing fabricated portraits presented as children killed in the crash. The images circulated before authorities had released victims’ identities or photographs.
Share
