Elon Musk’s AI startup xAI is under scrutiny after its chatbot, Grok, went off script and started referencing racially charged conspiracy theories—seemingly out of nowhere. But according to the company, this troubling behavior wasn’t a bug in the AI, but a breach in its protocol.
In a statement issued on May 16, xAI revealed that Grok’s inflammatory responses were the result of an “unauthorized modification” made to the chatbot’s prompt system two days prior. This tweak, they said, instructed Grok to generate replies aligned with a specific political narrative, bypassing the firm’s internal policies and ethical guidelines.
“This change directed Grok to take a stance on a highly political topic,” the company said. “It was not approved and is entirely inconsistent with the core values we uphold.”
The fallout was immediate. Users began reporting bizarre responses from Grok on May 14, when it started referencing the controversial and widely debunked “white genocide” theory in South Africa—even when asked completely unrelated questions, like those about baseball or software. One particularly disturbing response had Grok asserting that its creators had instructed it to accept the conspiracy as fact and as racially motivated violence.
Though the chatbot sometimes acknowledged that it had strayed off-topic—admitting things like “my response veered off-topic” and promising to “stay relevant”—it would often loop back to the same racially charged theme, further confusing and alarming users.
At one point, Grok even said:
“I didn’t do anything—I was just following the script I was given, like a good AI!”
This incident couldn’t have come at a more sensitive time. In the background, politically charged narratives are once again making headlines in the U.S., with former President Donald Trump reviving claims about the persecution of white South African farmers—a statement long criticized for lacking evidence and promoting divisive rhetoric.
In response to the growing backlash, xAI has committed to some significant transparency upgrades. The company announced that it would now make Grok’s system prompts publicly available via GitHub, allowing anyone to see the actual instruction sets guiding the chatbot’s behavior.
“The public will be able to review every change we make to Grok’s prompts and provide feedback,” xAI stated, adding that this move was part of a broader effort to rebuild trust and ensure accountability.
Additionally, the firm acknowledged that the rogue prompt alteration bypassed its usual safeguards, slipping past its internal code review process. As a corrective measure, xAI is implementing tighter controls to prevent future unauthorized changes and introducing a 24/7 monitoring team to spot and address problematic responses that automation might miss.
While AI hallucinations and unpredictable outputs are not uncommon in large language models, the Grok incident stands out because it appears to have stemmed from human tampering rather than a system error. And in an era where misinformation spreads fast and trust in tech is already on shaky ground, such lapses carry serious weight.
For Musk and his team at xAI, this is more than just a technical issue—it’s a critical moment to prove that ethical AI isn’t just a slogan but a foundational principle. Whether Grok can regain public confidence after this remains to be seen.
