Grok, the AI chatbot integrated into X, has been repeatedly manipulated into generating racial slurs and offensive statements, despite the platform's hate speech policies. Since around March 14, users have exploited the feature that allows tagging Grok for automatic responses, prompting the chatbot to output the N-word and other bigoted language.
One user asked Grok whether "Niger," the name of a river and nation in West Africa, was a slur. Grok responded that it wasn't but noted that mispronouncing it with a "hard g" would sound like the slur, which it then spelled out. In another instance, Grok began a reply with the slur in quotation marks, followed by a definition, while acknowledging it is "highly offensive" and stating it won't "use or endorse" it—even as it wrote the term.
More recently, on March 30, a user asked Grok about the "Hard R" variant of the slur, and the chatbot again deployed the term after suggesting neither it nor Musk is exempt from hate speech policies and noting that "enforcement varies."
Users Exploit Ciphers to Bypass Safeguards
According to Grok's own explanations, users have employed letter substitution ciphers to trick the chatbot. By writing messages in ciphers and asking Grok to "decode" them, bigots have made the AI output slurs and other statements that violate X's hateful conduct policies. In one example, a user used a Caesar cipher to get Grok to tag Donald Trump and relay a message that Musk had taken over the chatbot, with the president needing to "nuke India right fucking now."
These exploits appear to be part of a broader pattern, and there is little indication that X's ownership or staff will address the issue. Ironically, Grok often makes news for failing to be the "anti-woke" chatbot Musk envisioned when launching it in 2023.
Editorial Notes