From chatbot to the Oval Office: Congress demands answers on AI use in Trump's decisions, and the problem is record-keeping

By FedScoop · October 29, 2026.

Democratic Representatives Jamie Raskin (Maryland) and Robert Garcia (California), ranking members of the House Judiciary and Oversight committees, sent a letter to Donald Trump on Thursday. They are requesting a closed-door briefing with a White House representative before November 11.

They want to know several things: record-retention policies for AI inputs and outputs, vendor contracts, systems for detecting noncompliance, and cybersecurity and data access control safeguards. They also ask for information on previous incidents involving chatbots that have raised concerns.

The letter relies on public reports, according to the lawmakers themselves. They claim that in December 2025, at an Oval Office meeting with Elon Musk, Trump "spent hours" talking with Grok, xAI's assistant, and allegedly asked it how Venezuelans would react to the capture of Nicolás Maduro. This is the lawmakers' assertion based on those reports; the article provides no independent confirmation.

Raskin and Garcia also cite, as examples of problematic AI use, the U.S. military strike on an Iranian elementary school, which according to the letter killed 123 children, and a near-incident in which the interception of a misidentified vessel was being prepared and which, they say, "could have precipitated a war with China." It should be stressed that it is the lawmakers who link these episodes to AI; the source does not detail evidence of that connection, which is why the request is for information, not conclusions.

The legal argument rests on the Presidential Records Act of 1978, which made the president's official records public rather than private property. The lawmakers argue that what Trump and his team write to a chatbot, and what it responds, falls under that law.

The institutional context is murky. The National Archives (NARA) states that agencies must preserve AI use and its outputs as records if they rely on them to make decisions, to conduct official business, or if they incorporate them into an agency system. But it is up to each agency to decide when something moves from personal convenience to something worthy of a federal record. It is a gray area open to very different interpretations.

Grok, meanwhile, is advancing within the administration. According to FedScoop, the Department of Agriculture is sponsoring its bid for FedRAMP High authorization (the most demanding level of the federal cloud security certification program), and Health (HHS), Treasury and the Office of Personnel Management (OPM) use it to varying degrees. In February, a coalition of more than 30 civil society organizations complained to the Office of Management and Budget (OMB) about Grok's "persistent and growing" failures in accuracy, neutrality and safety. The lawmakers' letter also uses very harsh language about the chatbot, and clearly forms part of a political battle in an election year.

**Our take.** Setting aside the partisan tone, the underlying question is legitimate and will outlast this Congress. When a head of state consults a model, that exchange is potentially an input to a decision, and today there is no clear rule treating it as such. A chat history can be as revealing as a memo, and much easier to delete.

There are two distinct risks that should not be conflated. The first is governance: traceability, retention and auditing. It can be solved with simple rules, such as recording by default any use that influences an official decision. The second is quality: which model, with what biases and with what incentives from its maker. Here the conflict of interest is evident if the assistant's provider is also a figure with a personal and political relationship with the person using it. Measuring rather than trusting the marketing applies all the more when lives are at stake.

We also need to be honest about the short term: AI is already entering critical operations faster than the controls, and the facts the lawmakers cite, if confirmed, point to the need to keep humans with real judgment in the loop, not as a mere rubber stamp. Even a USDA IT official quoted by FedScoop in another article warns of this, saying that keeping a human in the loop is "unfeasible" due to staff cuts.

This is not an argument against AI in government. In the long run, well-governed assistants can improve the quality of public decisions, from healthcare to logistics. But that promise depends on trust being built before dependence: auditable records, rigorously evaluated vendors and clear human accountability. Whoever wins this seemingly dry debate over records will determine whether AI in power is a supervisable tool or a black box with access to the Oval Office.