What the magazine reports
TIME has published an account of a private December 2025 visit by Elon Musk to the White House in which, according to the magazine, President Trump spent hours putting questions to Grok, Musk’s chatbot, about his presidency and his legacy. The piece, written by Eric Cortellessa and published on 1 October, says one of those questions was how Venezuelans would respond if the United States captured Nicolás Maduro.
According to TIME, the chatbot replied that Maduro was a repressive and deeply unpopular dictator, and that many Venezuelans would probably celebrate his downfall. Trump ordered the operation to seize Maduro the following month, and when celebrations followed, the magazine reports, he came away impressed by the machine.
What it does and does not establish
It does not establish that Grok caused anything. The magazine does not report that the chatbot recommended the operation, and a president asking a question in December about an action he takes in January is not evidence that the answer decided it. The strikes on Venezuelan vessels were already under way when the conversation happened.
What it does establish, if the account is accurate, is the character of the use. A sitting president put a question of foreign-population sentiment to a commercial language model owned by a man who was at that point an adviser to him, in a private meeting, and was impressed when the output appeared to match events. That is a model being used as a source of political judgement rather than as a drafting or research tool, and it is the first reported instance of it at that level.
The pattern around it
Grok’s role in the American government is not new. TechCrunch, reporting on the TIME piece, noted that the Pentagon confirmed in June that the model had been used in connection with strikes during the Iran war. A separate federal chatbot launched this week runs on Gemini and Grok.
The obvious problem with the December exchange is that nobody can audit it. There is no published transcript, no record of how the question was framed, and no way to know what the model would have said to a differently worded prompt. A chatbot that tells a president what a foreign population thinks is making a claim no one checked, and the apparent confirmation — crowds in the street — is the kind of evidence that would also have followed a wrong answer delivered confidently.
What to watch
Whether the White House or xAI disputes the account; neither had commented in the reporting available on Thursday. Whether Congress asks what records exist of AI systems consulted in the run-up to military decisions, which is the question the account actually raises. And whether any of the three state and federal inquiries now open into frontier labs extends to how these models are used inside government, rather than only to what they do on their own.
