TIME Reports Trump Asked Grok for Hours
TIME cites officials present stating that in December 2025, President Donald Trump met privately with Elon Musk in the Oval Office, spending hours asking Grok questions about his presidency and legacy. This was about seven months after Musk left the Department of Efficiency.
The same source said he asked how Venezuelans would react if the U.S. captured Nicolás Maduro. Grok described Maduro as an unpopular dictator, and many might celebrate his ousting. The report did not prove that Grok suggested action, nor that the answers determined orders.
The U.S. executed the capture in Caracas on January 3 the following year. The source noted that celebrations in the streets led Trump to believe Grok was very smart. At the time of questioning, the government was already targeting Venezuelan vessels it claimed were involved in drug trafficking.
On September 29, the White House launched America.gov, where Trump attended and signed an executive order. National Chief Design Officer Joe Gebbia told CNBC that the backend uses Gemini and Grok, retrieving information from about 29,000 government websites, with answers limited to official sources. Google confirmed its involvement with Gemini, targeting over 100 million people needing public services.
The Department of Defense's enterprise platform GenAI.mil has integrated Grok, ChatGPT, and Gemini. Deputy Assistant Secretary James Mazol stated that out of about 3 million personnel, 1.7 million have used it, with about 500,000 using it almost daily, and plans to expand to classified and top-secret networks. Reports also indicated that the Pentagon's AI head mentioned Grok Gov was used during the June Iran conflict. The Department of Agriculture's internal deployment has not seen similar records.
The buyers are the presidential office, portals, and military users needing ready answers, while the sellers are xAI and concurrently Google and OpenAI. The event was driven by reports and the portal launch. The budget shifted from building small models to cutting-edge model seats, benefiting those entering the government directory, while suppliers not included in GenAI.mil and America.gov are under pressure.
Source: Public Information
ABAB AI Insight
Grok's entry into this scenario was through individuals, not through a bidding announcement. Even after Musk left the Department of Efficiency, he was able to bring the model to presidential inquiries in the Oval Office in December 2025. TIME's sources recorded the questions as reaction predictions, not operational plans. The celebrations after the capture on January 3 were used by the same source to explain why Trump found the answers useful.
The capital path is to first enter usage, then enter the directory. America.gov was launched by Joe Gebbia, co-founder of Airbnb and head of the National Design Studio, with Gemini and Grok running in parallel on the backend, retrieving from 29,000 government sites. GenAI.mil expanded from about 80,000 internal users to 1.7 million who have used it, with 500,000 high-frequency users, and the model list includes Grok, ChatGPT, and Gemini. The government is purchasing interchangeable cutting-edge models, not locking into a single supplier.
A comparable situation is when the White House earlier integrated commercial clouds into FedRAMP, and the Department of Defense included commercial software in enterprise accounts rather than developing everything in-house. The current position is one of control. Models have appeared at three levels: presidential inquiries, public portals, and military networks, but America.gov only returns official web pages, and the military version still needs to enter classified networks.
Structurally, this represents a transfer of pricing power. Policy Q&A and reaction judgments that used to be in staff memos can now first go through commercial models. If the directory continues to list multiple companies, none will have exclusivity; if the president's usage preferences are written into future procurements, the seats will lean towards the models that are named. The reports did not establish causality; capture decisions remain within the command chain, not in chat records.
ABAB News · Laws of Cognition
- First used for questioning, then appeared in the procurement directory.
- Listing three models in parallel does not yet constitute exclusive infrastructure.
- Predictions being remembered does not equate to commands being made by it.