Donald Trump reportedly spent hours talking to Grok during a private meeting with Elon Musk in December 2025, asking the model how Venezuelans would react to the capture of their own president. TechCrunch described the account on 1 October 2026, citing press reporting. The invasion followed on 3 January 2026.
Key takeaways
- December 2025: a multi-hour conversation with the model during a private meeting with Elon Musk
- The model reportedly called Maduro a "deeply unpopular dictator" whose downfall many Venezuelans would celebrate
- 3 January 2026: the US invasion of Venezuela and the capture of Nicolás Maduro
- June 2026: the Pentagon's AI chief confirmed Grok was used in operations during the Iran war
- Project Meridian: Musk, Palmer Luckey and Newt Gingrich are to advise the Pentagon on warfare technology
What is reporting and what is fact
Separating the layers matters here, because a press account and a confirmed fact carry very different weight in this story.
| Date | Event | Status |
|---|---|---|
| December 2025 | Trump's conversation with Grok during a meeting with Musk | press report |
| 3 January 2026 | US invasion of Venezuela and the capture of Nicolás Maduro | fact |
| June 2026 | Grok used in operations during the Iran war | confirmed by the Pentagon |
| 30 September 2026 | Project Meridian — Musk, Luckey and Gingrich as advisers | confirmed |
| 1 October 2026 | The December conversation described in the press | unconfirmed by xAI and the White House |
Grok's use in military operations was confirmed in June 2026 by the Pentagon's AI chief. The content of the December conversation, by contrast, comes from press reporting that neither xAI nor the White House has confirmed.
What the account omits matters too. It is unknown whether the model had access to intelligence data or answered from a publicly available corpus — the difference between an analytical tool and a generator of plausible text.
Why the praise is the problem
After the invasion Trump reportedly judged Grok "ingenious" because the prediction held. That is reasoning from a single trial: the model gave an answer matching what the questioner wanted and was rewarded for a hit indistinguishable from a guess. It is textbook confirmation bias?Confirmation bias: The tendency to weigh information that agrees with an existing belief more heavily, and to discount signals pointing the other way., amplified by the authority of a machine.
The scale of that mechanism is growing. Elon Musk's SpaceX and Palmer Luckey's Anduril hold Pentagon contracts worth up to $4.16 billion, and both men now co-lead a study of future military technology commissioned by Defense Secretary Pete Hegseth.
Why it matters
Language models have no calibration on geopolitics, because no training set of societal reactions to an invasion exists. They answer in a confident register regardless of grounding, and that is precisely the mode in which an existing conviction gets reinforced rather than tested.
What's next
- Neither xAI nor the White House has confirmed the content of the December conversation
- Project Meridian is to deliver recommendations to the Pentagon, with no date made public
- None of the study's three advisers has served in the US military
