House Democrats demand preservation of presidential AI chatbot logs

The demand for AI transparency

House Judiciary Committee Ranking Member Jamie Raskin and House Oversight Committee Ranking Member Robert Garcia sent a formal letter to the White House on October 28, 2026. They are demanding that the administration preserve all AI chatbot query logs, outputs, and model-response records. The lawmakers want to ensure that any interactions with AI during official duties are archived for the public record.

The letter specifically asks the White House to confirm if AI chatbots are being used in deliberative presidential decision-making. It also requires the administration to identify which specific systems are in use and by whom. The lawmakers have set a deadline of November 11, 2026, for the White House to provide a briefing on these practices.

This oversight effort comes as the administration has faced criticism for its handling of digital records. Because Raskin and Garcia are in the minority, they do not have subpoena power to force a response. However, the letter serves as a public pressure instrument to highlight what they describe as a lack of clear statutory frameworks for AI data.

The focus of this demand is anchored in reports that President Trump consulted Grok about Venezuela in December 2025. These consultations occurred weeks before U.S. forces took control of the country and detained Nicolás Maduro on January 3, 2026. Because no law currently requires anyone to save these specific types of queries, the information may have gone unpreserved.

The situation is complicated by the fact that Grok is a tool created by xAI, a firm owned and operated by Elon Musk. At the time of these consultations, Musk was acting as an unofficial advisor to the White House and leading the Department of Government Efficiency. This overlap has led to concerns regarding conflicts of interest in national security decisions.

The OECD's AI Incidents Monitor catalogued the Venezuela episode as a documented AI hazard. It noted that a tool from a firm with direct financial links to the White House was used as a policy-advisory tool. This occurred without a disclosed conflict-of-interest review or a framework for preserving the resulting records.

A legal void in record keeping

The Presidential Records Act (PRA) was signed into law in 1978 to ensure that official White House records are preserved. It covers documentary materials created or received by the President or their immediate staff in the course of official duties. However, the law was written for paper, tape recordings, and telegrams, not for modern large language models.

Legal scholars have noted that while the PRA's language is broad enough to possibly cover AI queries, no court has ruled on it. There is currently no enforcement mechanism to compel the preservation of chatbot logs. This creates a problem where records might exist on private or classified servers, making them effectively invisible to the National Archives.

The administration has further complicated this legal landscape by challenging the PRA itself. In April 2026, the Office of Legal Counsel issued an opinion stating that the PRA 'intrudes impermissibly on executive authority.' While this opinion does not dissolve the law, it signals that the administration may not feel obligated to comply with existing record-keeping rules.

Military and public AI use

The use of Grok extends beyond presidential consultations into active military operations. The Pentagon finalized a contract to deploy a classified variant called Grok Gov Model into the Maven Smart System. This system was used during Operation Epic Fury, which involved U.S. strikes on Iranian nuclear infrastructure in February 2026.

The integration of Grok into military platforms highlights the lack of a clear framework for archiving AI outputs in combat scenarios. There is no established rule for whether these outputs must be disclosed or archived. This lack of oversight is particularly notable given the high stakes of nuclear infrastructure strikes and military targeting.

Public-facing AI tools are also being modified to align with political goals. The America.gov AI chatbot was adjusted on its launch day to ensure its responses did not contradict the President's public positions. This demonstrates how AI-generated information can be altered in real time, creating another layer of complexity for historical record-keeping.

What comes next for AI oversight

The Raskin-Garcia letter is unlikely to receive a formal response given the administration's past reactions to similar demands. The White House has previously dismissed such requests as 'pathetic.' However, the letter establishes a documented timeline of the request and the administration's silence, which may be significant if political control of the House shifts.

Meanwhile, litigation continues regarding the constitutionality of the Presidential Records Act. While a preliminary injunction found the law likely constitutional, the administration's appeal is still pending. The specific question of whether AI logs count as 'documentary materials' has not yet been litigated in these proceedings.

Observers will be watching to see if any new regulations are proposed to bridge the gap between 1978 laws and 2026 technology. Without a clear statutory framework, the government remains in a state where AI-assisted policy decisions may leave no permanent trail for future historians or oversight bodies.