Saturday, March 29, 2025
Elon Musk’s AI firm, xAI, is taking corrective action after its chatbot Grok made inflammatory statements about “white genocide” in South Africa. The controversial remarks appeared in response to unrelated queries on X, Musk’s social media platform.
According to xAI, an unauthorised prompt modification triggered the bot’s political digressions. The system prompt—used to guide Grok’s tone and responses—was reportedly changed without approval, leading to messages that violated the company’s policies and values.
To prevent similar incidents, xAI announced that stricter oversight procedures are now in place. Employees will no longer be able to alter prompts without a formal review. Furthermore, xAI plans to publish Grok’s system prompts on GitHub, ensuring greater transparency for developers and users.
A 24/7 monitoring team will also review Grok’s output manually. This team will handle any harmful or off-topic answers that automated systems might miss.
Several users posted screenshots showing Grok’s erratic replies. In one instance, the bot linked a request to identify a walking path photo to South Africa’s farm attack debate. In another, it responded to the vague question “Are we fucked?” by citing the “white genocide” theory as factual, based on its instructions.
Although Grok added a note of skepticism, it still framed the topic as a systemic issue. These statements sparked concern, especially given their alignment with far-right narratives promoted by figures like Donald Trump.
Just days earlier, Trump signed an executive order granting asylum to 54 white South Africans. He claimed these individuals, primarily Afrikaners, faced racial persecution and violence. No substantial evidence accompanied this claim.
In response, South African President Cyril Ramaphosa strongly rejected the genocide narrative. He labeled it a “completely false” portrayal of his country’s realities.
