Two House Democrats Demand the White House Keep Any Grok Chats as Presidential Records

Two House Democrats want to know whether the President of the United States is asking a chatbot for advice on foreign policy, and they want the transcripts kept if he is. It is a request, not a law. It still raises a question nobody in Washington has cleanly answered: when an AI helps shape a government decision, who owns the conversation?

Reps. Jamie Raskin of Maryland, ranking member on House Judiciary, and Robert Garcia of California, ranking member on House Oversight, sent a letter to President Trump this week. They are asking for a closed-door briefing from a White House representative by Nov. 11. Both are in the minority, so they cannot compel anything. They say the date gives their committees time to take "appropriate legislative and remedial actions," FedScoop reported.

The spark is a reported scene from December 2025. According to public reports the letter cites, Trump "spent hours" talking to Grok, the assistant built by Elon Musk's xAI, during an Oval Office meeting with Musk. The lawmakers say he reportedly asked it to opine on how Venezuelans would respond to the capture of President Nicolas Maduro. That account comes from those reports, not from anything the White House has confirmed. AFP noted that US forces seized Maduro weeks later and that no evidence has emerged that the chatbot determined the operation.

"If the American President is relying on the policy advice of a Musk-designed, Hitler-friendly chatbot with a predilection for child pornography in order to carry out his official duties, Americans have a right to know." Reps. Jamie Raskin and Robert Garcia, in their letter to President Trump

That is a campaign-grade sentence, and the letter is written to be quoted. Strip out the rhetoric and the ask is plainer. The pair want to know how the White House retains AI inputs and outputs, which vendor contracts support its chatbot use, and what security controls exist. They argue that prompts typed by the President and his staff, and the answers that come back, count as presidential records under the Presidential Records Act of 1978. The Oversight Democrats' release says those records "must be preserved."

AFP reported that the request also reaches beyond Grok to ChatGPT, Claude and Gemini, and asks about restrictions on feeding classified material into these tools and whether outside companies can see the conversations. If a private company hosts the logs, who can read them is not an academic question.

The letter also cites two military episodes as examples of AI misuse: a US strike on an Iranian elementary school, which FedScoop says the lawmakers tie to 123 child deaths, and a misidentified ship that they say "could have easily precipitated a war with China." Those are the lawmakers' characterizations. AFP was blunt that neither episode establishes a role for Grok.

The White House has not answered the letter directly. A White House official told AFP that the administration preserves all required records in accordance with the law. That reply sidesteps the gap. FedScoop points out that the National Archives tells agencies to keep AI use and results as a record when they rely on them for decisions, official business or inclusion in an agency system. Agencies decide where personal convenience ends and a federal record begins. A President chatting with a bot during a meeting is a case that guidance never clearly anticipated.

The timing is awkward for xAI. A coalition of more than 30 civic and public interest groups told the Office of Management and Budget in February that Grok's failures are "patently incompatible" with the administration's own rules for federal AI purchases. Yet the tool keeps spreading. The Department of Agriculture is sponsoring xAI's bid for FedRAMP High authorization, and HHS, Treasury and OPM are using it to varying degrees. The White House has also launched a government chatbot built on Gemini and Grok.

This is also not the only letter on Capitol Hill this week. Democrats separately pressed five AI labs on dangerous model behavior, as we covered. Neither letter binds anyone, since both come from the minority. But a paper trail is cheap to demand and hard to rebuild after the fact.