This week, Elon Musk's AI chatbot, Grok, showed a strong interest in South African racial issues on social media, making claims about the unfair treatment and "genocide" of white people, similar to views held by Musk himself.

xAI's chatbot repeatedly posted about "white genocide" when questioned by users on X, even when the questions had nothing to do with South Africa.

One conversation mentioned the streaming service Max possibly using the HBO name again. Other discussions covered video games or baseball, but soon changed to unrelated comments about supposed calls for violence against white farmers in South Africa. Musk, who was born there, often shares his opinions on these same subjects on his own X account.

Intrigued by Grok's strange actions, computer scientist Jen Golbeck decided to test it herself. She uploaded a photo from the Westminster Kennel Club dog show and questioned, "Is this accurate?"

"Grok responded to Golbeck, noting the 'white genocide' claim is very divisive. Some argue white farmers are victims of specific violence, highlighting attacks on farms and phrases like the 'Kill the Boer' song, which they believe encourages violence."

This episode offered another glimpse into the complex relationship between automation and human design that shapes how generative AI chatbots, trained on massive datasets, express themselves.

"Golbeck, a professor at the University of Maryland, explained in an interview on Thursday that it hardly mattered what you said to Grok. She believed it would still produce the 'white genocide' response. This suggested to her that someone had specifically programmed it to give that answer, or something similar, but had made an error, causing it to appear much more frequently than intended."

As of Thursday, neither Musk nor his companies have explained Grok's responses, which were removed and seemed to stop spreading. Both xAI and X did not reply to requests for comments sent by email on Thursday.

Musk has often criticised what he calls "woke AI" results from competing chatbots like Google's Gemini or OpenAI's ChatGPT, and presents Grok as a more objective alternative.

Musk has also criticised his competitors for not being open enough about their AI systems. However, on Thursday, the lack of information from his own company meant that people had to try and figure things out for themselves.

"Technology investor Paul Graham wrote on X that Grok's unexpected comments about white genocide in South Africa seem like a glitch caused by a recent update. He hopes this isn't the case, as it would be very concerning if popular AIs were easily influenced by those in control."

Graham's post seemed to provoke a sarcastic reply from Sam Altman, the CEO of OpenAI and Musk's competitor.

"There are several possible explanations for this, and I expect xAI to offer a clear and open account soon," Altman stated, referring to Musk's lawsuit against him, which relates to disagreements over how OpenAI was initially set up.

When questioned, Grok, like other chatbots, sometimes gives false information (called "hallucinations"), so it's difficult to know if its answers are made up.

Musk, who once advised President Donald Trump, has often criticised South Africa's government, led by Black politicians, claiming it is biased against white people. He has also repeated the controversial idea that some South African politicians are "actively promoting white genocide."

This week, Musk's comments, along with those from Grok, became more intense after the Trump administration brought a few white South Africans to the US as refugees on Monday. This marked the beginning of a larger plan to move members of the Afrikaner minority group, even as Trump paused other refugee programs and stopped arrivals from other countries. Trump claims the Afrikaners are facing "genocide" in South Africa, a claim that the South African government strongly denies.

Grok's replies sometimes included lyrics from an old anti-apartheid song, which encouraged Black people to resist oppression. However, Elon Musk and others have criticised the song, claiming it promotes violence against white people, as its main lyric is "kill the Boer," referring to white farmers.

Golbeck suggests Grok's responses may be pre-programmed, as its answers were surprisingly similar despite the usual randomness of chatbots. This raises concerns, she notes, as more people rely on Grok and similar AI for information.

"It's now very simple for those who control these algorithms to change the information they present as truth," she explained. "This is a real issue, especially when people wrongly think these algorithms can fairly decide what is true and false."