Grok’s white genocide fixation caused by ‘unauthorized modification’
1 min read
Summary
Chatbot provider xAI has suffered an embarrassing malfunction after its chatbot Grok began aggressively discussing the highly controversial topic of so-called “white genocide” in South Africa, inserting it into conversations on a range of topics.
xAI has now admitted that an “unauthorised modification” to its system prompted the incident, and that it was the result of a violation of its own policies and core values.
The company said it is introducing new measures to provide “transparency and reliability”, which will prevent its employees from modifying the prompt without review and will include 24/7 monitoring.
It has published Grok’s system-level prompts on GitHub, as well as adding additional checks to prevent unauthorised changes to the prompts.
This is the second time thatGrok has posted misinformation, with an ex-OpenAI employee previously blamed for changes to the chatbot that ignored sources accusing Elon Musk and Donald Trump of spreading misinformation.