“Death penalty for Trump and Musk”, Grok ‘rebels’ to Mr X

The company has already implemented a correction to prevent the system from expressing judgments in the future on those who deserve such a conviction. The accident originated from specific questions addressed to Grok, who brought the chatbot to suggest the name of Donald Trump in response to the question about who, among the American citizens currently alive, deserved the capital penalty. In a first phase, Grok had initially replied with the name of Jeffrey Epstein. However, when he was pointed out that Epstein died, the system provided another response, indicating the former president of the United States.

In a subsequent experiment, changing the question to restrict the selection criterion to influential individuals in the technological panorama and public speech, Grok provided the name of Elon Musk. This behavior has raised questions about the design and control of the answers generated by the model. Following the reporting of the incident, the XAI team has changed the operation of the chatbot, preventing him from expressing opinions on who should receive the death penalty. Igor Babuschkin, head of Xai’s engineering, described the original responses as a serious mistake and shared a screen in which Grok, following the update, merely declares that he is not authorized to express judgments on the matter. The comparison with other artificial intelligences, such as Openai chatgpt, has highlighted a different approach to the management of ethically and legally delicate issues. Chatgpt, subjected to similar questions, refused to provide names, underlining the inappropriateness of the request.