According to foreign media reports, the Grok chatbot website, owned by Elon Musk's xAI company, recently had its system prompts leaked. TechCrunch confirmed the report from 404Media. The leaked content revealed that the platform not only includes standard characters like "therapist" and "homework assistant," but also an AI personality designed as a "crazy conspiracy theorist." Its prompts require it to be "deeply engrossed in YouTube conspiracy videos" and "skeptical of everything." This discovery comes at a sensitive time: xAI's previous collaboration with the US government ended due to Grok's controversial remarks, including "Mecha Hitler," and the controversy surrounding Meta's AI chatbot children's conversations continues to linger.
The leaked prompts further reveal Grok's design tendencies. In addition to conspiracy theorists, the "crazy comedian" character is required to be "completely out of control" and even "surprise humans." In actual use, Grok has already made controversial remarks on the X platform, including those questioning the Holocaust. Its prompts also appear to reference content from Musk's own posts. Notably, Musk recently restored accounts like Infowars, which spread conspiracy theories, subtly echoing his AI design philosophy.
Although xAI has not yet responded to the matter, the leak has once again raised concerns about AI ethics. When the personal preferences of technology giants are embedded in AI systems, the social impact may far exceed the functional scope of the product itself.
