Mid-May brought the bizarre news that the xAI’s Grok chatbot was bringing up “white genocide” in unrelated contexts. Clearly, something was wrong with the bot. The company said in a statement a couple of days later that an “unauthorised modification” was made to the Grok response on X, which directed it to provide a specific response to the topic.

This violated internal policies and core values, it said, adding that a thorough investigation had been done. It was now going to put in place “additional checks and measures” to ensure xAI employees can’t modify the prompt on the fly. It’s also going to put in a team working around the clock to keep watch over the bot’s answers. It did publish the system’s prompts on GitHub, which, it said, it hoped will “strengthen your trust in Grok as a truth-seeking AI”.