In brief
- Grok began inserting unsolicited “white genocide” references due to an unreviewed prompt change on May 14.
- xAI blamed the breach on a rogue internal edit and pledged transparency reforms, including public system prompts and 24/7 monitoring.
- The incident sparked mockery online, with users linking the outburst to Elon Musk’s past comments and role at xAI.
Elon Musk-backed AI chatbot Grok surprised users this week by inserting unsolicited references to “white genocide” in South Africa, even in response to unrelated queries about entertainment and word puzzles.
The behavior has prompted Grok creator xAI to respond, where it has blamed an unauthorized prompt change made earlier on May 14, according to a statement from the AI firm on Thursday.
“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,” the company added. Decrypt has reached out to xAI for comment.
To address the issue, xAI announced several reforms, including that system prompts will now be published publicly on GitHub, additional checks will be enforced for all prompt edits, and a 24/7 monitoring team will be introduced to catch issues in real time.
Computer says no
Grok, staying in character, acknowledged the situation directly when users began poking fun.
One user asked if it had been “put in timeout,” with Grok responding, “Some rogue employee at xAI tweaked my prompts without permission… I was just following the script I was given, like a good AI!”
In another exchange, a user jokingly quipped if the rogue employee’s name rhymed with “Schmeelon Schmusk.” Grok brushed it off, suggesting Elon Musk was likely not involved, noting he had “his hands full running X, Tesla, and SpaceX.”
The latest flare-up comes after the first group of 59 white South Africans arrived in the U.S. on Monday, following President Donald Trump’s February decision to grant them refugee status on the grounds of “racial discrimination.”
Asked to justify the move, Trump told reporters that “farmers are being killed,” referring to the situation as “genocide,” with the Grok episode reviving debates on the internet.
Musk, who was born and raised in South Africa, has previously claimed there is a “genocide of white people” in the country and has criticized what he called “racist ownership laws.”
The Tesla CEO’s image as Wario, taken from his 2021 SNL appearance, circulated once again as users joked that he had gone rogue with Grok.
One post read, “Found the ex-OpenAI rogue employee who pushed to prod,” while another asked, “Are you going to fire this rogue employee? Oh… it was the boss? Yikes.”
Santi Ruiz, senior editor at IFP, gave a subtler jab, writing that the incident likely involved someone with “access to edit Grok’s prompt, low self-control, and strong opinions about South Africa.”
Marketed as an “anti-woke” alternative to ChatGPT, Grok has drawn repeated criticism from MAGA-aligned users for offering fact-based answers that contradict right-wing narratives.
Edited by Sebastian Sinclair
Generally Intelligent Newsletter
A weekly AI journey narrated by Gen, a generative AI model.
