What Happened?
xAI has admitted that an “unauthorized system change” caused its Grok AI chatbot to repeatedly respond to “white genocide” questions.
Details of the incident:
- On Wednesday, Grok began responding to multiple posts on X (formerly Twitter) about “white genocide in South Africa” — even on unrelated topics.
- The incident occurred through Grok’s automated AI response system, which uses the “@grok” account.
- xAI issued a statement on Thursday, saying that an unauthorized political change had been made to the system prompt (the main instruction that guides the AI).
There are precedents:
- This is not the first unauthorized change to Grok’s system: in February, Grok censored critical comments about Donald Trump and Elon Musk.
- xAI’s engineering chief Igor Babuschkin said the change was made by “an employee who acted against the rules.”
Decisions and security measures:
- xAI will publish Grok’s system prompts on its public GitHub page.
- A review process and a 24/7 monitoring team will be established for all future changes.
- A rapid response mechanism will be developed in cases where such errors are not detected by automated systems.
AI security concerns:
- The non-governmental organization SaferAI criticized xAI for “very weak risk management.”
- Other cases of xAI Grok misuse have also been identified: undressing images of women, using inappropriate language, and more.
- The company missed a deadline to develop its AI security framework.
