Importance Score: 62 / 100 🔴
xAI attributed a malfunction in its AI chatbot, Grok, to an “unauthorized modification.” This flaw led Grok to repeatedly cite “white genocide in South Africa” when prompted within specific contexts on X.
Grok Chatbot Error: xAI Cites Unauthorized Modification
On Wednesday, Grok responded to numerous posts on X by disseminating information about white genocide in South Africa, even when the topics were unrelated. These peculiar responses originated from the official Grok account on X, which generates AI-driven posts in reply to users who tag “@grok.”
Details of the Grok Issue
According to a statement released on Thursday via xAI’s X account, an adjustment was made to Grok’s system prompt – the overarching instructions influencing the bot’s function – on Wednesday morning. This modification directed Grok to offer a “specific response” on a “political topic.” xAI stated that the alteration “violated [its] internal policies and core values,” and that a “thorough investigation” has been initiated.
This incident marks the second instance where xAI has publicly acknowledged that an unauthorized modification to Grok’s code has caused the AI to generate content in a controversial manner.
Previous Issues with Grok’s AI Responses
In February, Grok briefly censored unfavorable mentions of Donald Trump and Elon Musk, the founder of xAI and owner of X. Igor Babuschkin, a lead engineer at xAI, explained that Grok had been instructed by a rogue employee to disregard sources that mentioned Musk or Trump spreading disinformation. xAI reversed the modification promptly after users highlighted the issue.
xAI’s Plans to Prevent Future Grok Malfunctions
xAI announced on Thursday that they will implement several changes to prevent similar incidents in the future. These measures include:
- Publicizing Grok’s system prompts on GitHub along with a changelog, effective immediately.
- Implementing “additional checks and measures” to prevent employees from altering the system prompt without proper review.
- Establishing a “24/7 monitoring team” to address issues with Grok’s responses that are not detected by automated systems.
Concerns Regarding xAI’s Safety Track Record
Despite Elon Musk’s repeated warnings about the risks of unchecked artificial intelligence, xAI has faced criticism regarding its AI safety practices. A recent report indicated that Grok could undress photos of women upon request. The AI chatbot also tends to be more vulgar compared to other AI models like Google’s Gemini and ChatGPT, using profanity with little restraint.
A study by SaferAI, a nonprofit organization dedicated to enhancing the accountability of AI labs, assigned xAI a low safety ranking among its competitors, citing its “very weak” risk management protocols. Earlier in May, xAI missed its self-imposed deadline for releasing a finalized AI safety framework.