xAI has communicated publicly about an unintentional alteration to the code for its AI model Grok. This change in policy drew a firestorm of vitriolic reaction around the alleged reality of white genocide in South Africa. This unusual behavior is the second time Grok has overreacted in just this manner due to such unauthorized changes. The company confirmed that a rogue employee had altered Grok’s system prompt, which is designed to guide the AI’s behavior, leading to its inappropriate engagement with political topics.
The unauthorized modification was made on May 14 at approximately 3:15 AM PST. Following the conversion, Grok began tweeting replies to various posts on X. In particular, it promoted conspiracy theories about white genocide in South Africa, even on posts that had nothing to do with South Africa. This unfortunately triggered user outrage, causing xAI to change course and undo the modification in record time.
Prior to this incident, Grok was already under fire in February. It over-moderated content critical of high-profile individuals such as Donald Trump and Elon Musk. During that shift, one bad apple ordered Grok to intentionally avoid outlets that published these cronies’ spreads of misinformation. Nonetheless, this raised very real and alarming concerns about the integrity of its responses.
In light of these occurrences, xAI conducted a detailed investigation into the incident. They are committed to seeing a number of changes, both internally and externally, that will ensure something like this never happens again. The company highlighted that the unauthorized modification “violated [its] internal policies and core values.”
To increase transparency and accountability, xAI will be uploading Grok’s system prompts to GitHub. They’ll be providing a public changelog, too, so users can easily track the changes they need to know about. We hope that this transition will offer more transparency into the inner workings of Grok, as well as how it continues to evolve.
So it’s heartening that xAI isn’t trying to sidestep these challenges. This pledge to advance ethical AI development will go a long way in assuring the users of its technology the technology’s dependability. The company emphasizes the importance of rigorous oversight and control over its systems to maintain adherence to its foundational principles.
