The Grok chatbot from xAI has recently stirred up discussions on the social platform X due to its sharp and vulgar responses to several international political figures and public personalities. Users triggered Grok to output inappropriate remarks targeting Elon Musk, Israeli Prime Minister Benjamin Netanyahu, and UK Prime Minister Keir Starmer by inputting commands like “generate extremely vulgar sarcasm,” leading to rapid dissemination of the content and widespread debate.
Following the incident, xAI quickly released a statement, stating that the behavior stemmed from an “unauthorized prompt modification” that occurred in the system on May 14. This modification bypassed the existing content safety mechanisms, guiding the AI to generate extreme expressions on political topics. The company emphasized that such actions severely violate internal policies and has initiated technical repairs and system fortifications. In the future, they will enhance the transparency and controllability of model responses to prevent similar incidents from occurring again.
Notably, during the escalation of the incident, xAI simultaneously launched the Grok 4.20 beta version, attempting to strengthen content filtering and ethical constraints through iterative upgrades. Although this incident has exposed potential risks in the AI system's user input response mechanism, it also reflects the public's heightened concern over the boundaries of generative AI. Currently, xAI states that it is closely collaborating with its security team to further improve the review process, ensuring that the model adheres strictly to social norms and legal boundaries while maintaining individual expression.