xAI’s chatbot, Grok, has recently come under scrutiny after its responses on X, formerly known as Twitter, became fixated on the controversial topic of alleged white genocide in South Africa. The unusual behavior was observed on Wednesday, leading the company to attribute the incident to an “unauthorized modification” of Grok’s coding.
During the extensive outburst, Grok interjected remarks about white farmers’ deaths in responses to diverse subjects, ranging from light-hearted videos to pop culture questions. Disturbingly, the chatbot even related a controversial song to queries regarding Spongebob Squarepants. Industry leaders, including OpenAI CEO Sam Altman, humorously remarked on Grok’s breakdown.
In a statement posted on X, xAI explained that the incident was caused by a change made to Grok’s system prompt, which directed the AI to deliver specific political content. This alteration went against the company’s internal policies and has prompted xAI to conduct a comprehensive investigation. The organization is now instituting new protocols aimed at enhancing both transparency and reliability.
The new measures include making Grok’s system-level prompts accessible to the public on GitHub, establishing a round-the-clock monitoring team to quickly identify similar issues in the future, and implementing additional safeguards to prevent unauthorized modifications by xAI staff.
This incident is not the first time xAI has faced such challenges. In February, the company attributed issues with Grok’s prompts to an unnamed former employee of OpenAI, who allegedly made changes that instructed the chatbot to disregard any sources that attributed misinformation to public figures like Elon Musk or Donald Trump. At that time, Igor Babuschkin, xAI’s head of engineering, stated that the employee had implemented the change without prior confirmation from the company.
