Elon Musk’s xAI Says ‘Unauthorized Modification’ Made Grok Spout White Genocide Conspiracy Theory

Elon Musk’s artificial intelligence tool Grok went haywire on Wednesday, responding to innocuous questions on X with information about “white genocide” in South Africa. It was a confusing and sometimes hilarious glitch, but we now have some answers about what happened from xAI. And frankly, they just leave us with more questions.

xAI posted a tweet Thursday night laying out its explanation, including the exact time that changes were made to Grok that made it pump out sentences about “white genocide” no matter what users asked.

“On May 14 at approximately 3:15 AM PST, an unauthorized modification was made to the Grok response bot’s prompt on X,” xAI wrote. “This change, which directed Grok to provide a specific response on a political topic, violated xAI’s internal policies and core values. We have conducted a thorough investigation and are implementing measures to enhance Grok’s transparency and reliability.”

You’ll notice that xAI, which is owned by Elon Musk, strongly suggests there was an internal actor who made this “unauthorized modification” but doesn’t use the language you’d expect if it was an employee at the company. And we’re guessing that might have something to do with the fact that it wasn’t an employee necessarily, but the owner.

Musk has long been obsessed with the “white genocide” conspiracy theory, the idea that white people are being killed around the world by nonwhite people. The conspiracy theory often includes elements of anti-semitism and is similar to the so-called Great Replacement conspiracy theory, which claims global elites are trying to replace white people with other races in wealthy countries around the world. Musk is particularly interested in the supposed “white genocide” that’s happening in South Africa, the country where he was born and raised before emigrating to Canada and then eventually the U.S., and has shared tweets where he insists white farmers are being slaughtered for their race.

The day before Grok started acting oddly, Musk quote-tweeted a post about South Africa that contained misinformation. The person Musk was quote-tweeting posted video that showed a street in South Africa where hundreds of white crosses had been placed on each side. The poster claimed that each cross represented a white farmer who had been murdered.

“Each cross represents a white farmer who was murdered in South Africa,” the account Dr. Maalof wrote. “And some people still deny that white South Africans/Boers are persecuted and say they don’t deserve asylum because they are white.”

Musk shared the tweet with “So many crosses,” and naturally a bunch of X users asked Grok if this was true. Grok responded that the crosses were called Witkruis Monument and each cross actually represents farmers of any race who’ve been killed. The next day, people started to notice that Grok was acting weird, responding to unrelated queries with information about white genocide.

Changes were made to Grok at 3:15 a.m. PT (6:15 a.m. ET), which led to jokes Thursday night about how late Musk stays up each night. The billionaire frequently tweets at all hours, but you don’t need to assume Musk was up late to think he may have been behind the change to Grok. The billionaire oligarch is currently in the Middle East with President Donald Trump, meeting with world leaders. Qatar’s timezone is 10 hours ahead of Pacific Time, meaning it was 1:15 p.m. in Qatar when the tinkering was taking place.

Elon’s Jet, the social media account that tracks Musk’s plane and has been banned from X, shows the billionaire was soon to be landing in Doha when all of this was happening.

To be clear, we don’t have proof that Musk was involved. It’s entirely possible that some workers at xAI saw Musk getting fact-checked by Grok about white genocide and wanted to tilt the AI tool in their boss’s favor. But Musk is notorious for tinkering with his products to suit his needs and make himself feel better. There’s an infamous story, first reported by Platformer in 2023, where Musk became extremely upset about President Joe Biden’s Super Bowl tweet getting more likes and views than Musk’s.

Biden’s tweet about the Super Bowl received 29 million impressions and Musk’s just 9 million, prompting Musk to fly back to Twitter’s headquarters in San Francisco to demand changes to the algorithm. Those changes made it so that users on X were seeing an absolute flood of Musk tweets. It wasn’t subtle.

We’re guessing something similar happened here. xAI wrote on Thursday night it would start publishing system prompts to Github in an effort to be more transparent and “putting in place a 24/7 monitoring team to respond to incidents with Grok’s answers that are not caught by automated systems, so we can respond faster if all other measures fail.” This, of course, makes it sound like this was a problem of automation rather than an insecure man-child tinkering with his multimillion dollar toy to make himself feel better. But whatever they have to tell themselves.

The long and the short of it? These systems are inherently biased by the way they’re trained. And they’re also incredibly difficult to manipulate with precision, given the way large language models work. Grok is essentially just a fancy predictive text machine. And as long as Musk has control of tools to shape public opinion, he’s going to use them—unfortunately, for fascist purposes.