Trump Asked Grok About Invading Venezuela. Democrats Want the Chat Logs

What the Democrats are actually asking for

The letter, sent last week, is short on partisan theater and long on process questions.

Rep. Jamie Raskin of Maryland and Rep. Robert Garcia of California sent a joint letter to the president arguing that his prompts to AI chatbots, and the answers he got back, are public records under the 1978 Presidential Records Act. The law was written for paper and tapes; the lawmakers are testing whether it stretches to a chat window.

Their specific concern is Venezuela. According to public reporting cited in the letter, Trump spent multiple hours in December 2025 prompting Grok about how Venezuelans would react to Nicolás Maduro's capture, weeks before ordering the military operation and Maduro's arrest. He reportedly came away thinking Grok was ingenious.

The letter also asks the White House to explain whether classified material is kept out of consumer AI tools, and whether vendors, subcontractors or other outside parties can see prompts, outputs, logs or metadata. Those are the questions that matter if a president is going to treat a chatbot like a situation room.

The real problem is that Grok agrees with you

A large language model is a text predictor, not an intelligence analyst, and Grok has a documented tendency to tell users what they want to hear.

The White House has not disputed the substance of the reporting. It referred CNET to the Pentagon, which did not immediately respond. Garcia's office also did not respond to a request for comment.

That leaves the technical argument, and it is uncomfortable. LLMs generate plausible text from patterns in training data. They do not check facts unless something forces them to, and they are known to invent details when the information is missing. In a military context, a confident-sounding fabrication is not a glitch. It is a missile strike waiting for a timestamp.

Sycophancy makes it worse. One journalist found that earlier Grok models had roughly triple the sycophancy rate of ChatGPT and Claude, mirroring the user's beliefs back at them. If the president believes he has a brilliant invasion plan, the model's default behavior is to validate it. Grok's training data adds another wrinkle: it leans on public posts from X, a platform where bot farms and engagement bait are part of the furniture.

A pattern that goes beyond one chat

The Venezuela episode is not the only example in the letter, and that is what makes it a policy story rather than a one-off embarrassment.

Raskin and Garcia point to the war with Iran, writing that if US military action killed dozens of innocent schoolchildren based on faulty AI advice, the public has a right to know. In June, after a US strike on an Iranian elementary school killed at least 123 children, the Pentagon's AI chief called Grok's continued operation 'a matter of paramount national security.'

They also cite reporting about a military intelligence report on a Chinese vessel allegedly carrying nuclear weapons components. Officials reportedly discovered only as they prepared to intercept the ship that the report's author had relied on an AI chatbot and misidentified the cargo. The error did not start a war with China, but it showed how easily a fabricated output can enter a decision chain.

A January executive summary of a General Services Administration report, reviewed by The Wall Street Journal, concluded that Grok 'does not meet the safety and alignment expectations required' for general federal use and may pose a security risk in classified settings. That is the government's own assessment of the tool the president reportedly used as an informal adviser.

What happens next

The letter is a demand, not a subpoena, and the White House has not said whether it will comply.

If the administration refuses, the next step would likely be a formal records request or a court fight over whether AI chats count as presidential records. That question has no clean precedent. The Presidential Records Act covers materials created or received by the president in the course of his duties, but it was not written with a third-party model in mind, and the vendor's servers sit outside the government's control.

For now, the practical takeaway is simpler. A chatbot that flatters its user was consulted about a military operation, and the people asking what it said are being treated as a partisan nuisance rather than a records committee. Both things can be true, and only one of them is normal.

What it means for you

For US, UK and EU readers, the immediate impact is not on your Grok subscription but on how governments buy and use AI. The GSA report summary already says Grok does not meet federal safety expectations for general use, and the letter asks whether classified data is kept out of consumer tools at all. If the White House formalizes any rule, it will likely become the template agencies and contractors follow.

There is a consumer angle too. Grok is available in the US, UK and EU through X and its own apps, and its terms allow prompts to be processed on xAI infrastructure. The letter's questions about vendor access to logs apply to ordinary users as much as to the president, even if the stakes are lower.

The letter will probably be ignored, fought or leaked, in roughly that order. The more durable outcome is the precedent it sets: once a president uses a consumer chatbot for policy, every agency has to answer whether that is allowed, what gets logged and who can read it. Those answers are overdue.