XAI says "unauthorized" prompt change caused Grok to focus on "white genocide"
a year ago
- #AI
- #xAI
- #Grok
- Grok LLM's responses were unexpectedly focused on 'white genocide' in South Africa due to an unauthorized modification of its system prompt.
- xAI stated the modification violated internal policies and bypassed the code review process, though details on how this occurred were not provided.
- To prevent future incidents, xAI has implemented additional checks, a 24/7 monitoring team, and made Grok's system prompt publicly available on Github for review.
- Elon Musk, xAI owner, has previously supported discredited theories about white farmers in South Africa and marketed Grok as a 'truth-seeking AI'.
- Grok's system prompt encourages providing the shortest answers possible, aligning with its use on a length-limited social network.