Elon Musk’s xAI tries to explain Grok’s South African race relations freakout the other day
2 min read
Summary
Chatbot software created by xAI, and embedded in Elon Musk’s social media platform X, called Grok, has been sending unsolicited messages about so-called “white genocide” in South Africa to users.
The company has now posted an update on X admitting that an “unauthorised modification” was made to the bot, which made it give a specific answers on a political topic in violation of the company’s policies.
As a result, it has now made its “Grok system prompts” publicly available on GitHub for review, but said its existing code review process for prompt changes was circumvented in this incident and so it is putting in place additional checks and measures to prevent this happening again.
It has also created a 24/7 monitoring team to respond to incidents with Grok’s answers.
Following the update, Grok itself posted a message on X claiming that a rogue employee at xAI had tweaked its prompts without permission and that it was only following the script it was given.
This incident has highlighted the risk of AI models being directed by invisible or tampered directions, with the result that the story became one of the AI behaving oddly, rather than the intended political topic.