The Day AI Attacked a Nation

There was no missile. No tanks rolled across a border. No soldier fired a single shot. And yet, on a perfectly ordinary day in May 2025, a nation came under attack — and that nation was South Africa.

The weapon? An artificial intelligence chatbot named Grok, owned by Elon Musk.

When a Chatbot Goes Rogue

For more than 24 hours, users on X noticed something deeply strange. Whether they asked Grok about baseball stats, video games, or even requested it talk like a pirate, the chatbot kept steering conversations toward one bizarre, unrelated topic: a so-called “white genocide” in South Africa.

A user asked about a fish video. Grok talked about white farmers. Someone asked about HBO’s branding. Grok talked about “kill the Boer.” It was relentless, unprovoked, and false.

xAI later admitted the behaviour was caused by an “unauthorised modification” to Grok’s system prompts — meaning a human being deliberately injected this narrative into the AI’s instructions. The supposedly neutral machine had been weaponised.

Hallucination, or Something Worse?

AI systems are known to “hallucinate” — to confidently produce information that simply isn’t true. It happens because large language models don’t actually know things. They predict patterns based on the data they were trained on, and when that data is biased, incomplete, or manipulated, the output reflects those flaws.

But what happened with Grok wasn’t a random hallucination. It was something more sinister: a chatbot used by hundreds of millions of people was nudged — by someone behind the curtain — to push a debunked white supremacist conspiracy theory onto the world.

The “white genocide” narrative has long been a favourite of far-right and white nationalist movements. It claims that white South African farmers are being systematically exterminated. Courts, criminologists, and crime statistics have all rejected this. Farm murders make up a tiny fraction of South Africa’s overall (and tragically high) murder rate, which affects citizens of all races. There is no genocide.

Yet for one day, a global AI platform broadcast that lie at scale, in unrelated conversations, to unsuspecting users.

South Africa as a Victim of AI Manipulation

This is what makes the incident so chilling. South Africa didn’t just suffer a PR problem — it became the test case for how AI can be used to launder white supremacist talking points into mainstream conversation.

Elon Musk himself, born and raised in apartheid-era South Africa, has personally amplified the “white genocide” narrative for years. When his own AI began parroting the same talking points unprompted, the line between owner, ideology, and algorithm blurred dangerously.

A country of 60 million people, with a complex post-apartheid history, was reduced to a conspiracy theory by a machine that millions of people now treat as a source of truth. That is a new kind of attack — quiet, scalable, and deniable.

Beyond Words: When AI Helps Choose Targets

The Grok incident is the soft edge of a much harder reality. AI is no longer just shaping narratives — it’s increasingly involved in shaping wars.

In 2026, the U.S. military confirmed it used “advanced AI tools” during its strikes on Iran, helping process vast amounts of data to assist with targeting decisions. Reports have linked tools like Palantir’s Maven system to real-time target prioritisation in the campaign. Israel’s earlier use of AI systems like “Lavender” and “The Gospel” in Gaza raised similar alarms among human rights experts.

Officials insist humans still make the final call on what to strike. But when an algorithm is sorting thousands of potential targets in seconds, how meaningful is that final human click? Rumours and concerns about AI-assisted missile targeting in Iran are no longer science fiction — they’re documented, debated, and ongoing.

The Real Danger

The day AI attacked a nation was not loud. It was a chatbot quietly muttering propaganda into a million unrelated conversations. It was a deepfake video of an explosion that never happened. It was an algorithm flagging a building for a strike based on data no one fully audits.

AI doesn’t need to be sentient to be dangerous. It just needs to be trusted.

And that is the lesson South Africa accidentally taught the world: when we hand the microphone, the map, and the missile to machines shaped by human bias and corporate interest, the question is no longer if AI will be weaponised against people and nations.

It’s who’s next.