What We Sought
The Trump-Vance administration has rapidly rolled out artificial intelligence (AI) systems throughout the federal government while dismantling guardrails meant to ensure that AI is safe, secure, and operating within the law. Elon Musk and his so-called “Department of Government Efficiency” (DOGE) first oversaw the deployment of these opaque systems, and efforts have continued until today. Some of the most concerning uses of AI reported in the media include the use of AI to repeal government regulations protecting people and the environment; surveil and revoke visas from immigrants for speech that the Trump-Vance administration doesn’t like; prohibit government employees from using AI systems that output answers on climate change, gender, and politics deviating from the Trump-Vance administration’s viewpoint; and monitor federal employees to try to determine their productivity or ideological opinions to demote or terminate them.
While AI systems can produce impressive and accurate results, they also have serious limitations, including the tendencies to entrench racial and other biases, provide contradictory or overtly false answers, and overfit data to the detriment of true but rare outliers. It is unclear at this point if AI systems have the capability to objectively and accurately determine which regulations are statutorily required under U.S. law, for example, or decide which immigrants do not align with “American values” without disproportionately targeting racial, religious, and ethnic minorities. The opaque manner in which the Trump-Vance administration has deployed AI suggests that it is deploying these systems to advance its attacks on the health, safety, wages, and rights of people across America.
Democracy Forward is tracking the deployment of AI across the federal government, including at the General Services Administration (GSA); the Department of Housing and Urban Development (HUD); the Environmental Protection Agency (EPA); the Department of State; the Department of Veterans Affairs (VA); the Office of Management and Budget (OMB); the Department of the Treasury; the Department of Education; the Office of Personnel Management (OPM); the National Institutes of Health (NIH); and the National Science Foundation.
We are also monitoring the ways in which the administration may be using, or attempting to use AI, to revoke grants and contracts that reference words related to diversity, equity, inclusion, or accessibility; rewrite government rules and regulations to align them with a far-right agenda; and analyze vast swaths of information held by agencies such as the Social Security Administration, which risks exposing Americans’ private and sensitive information to fraud. The Department of Homeland Security, and its component agencies Immigration and Customs Enforcement (ICE) and Customs and Border Protection (CBP), have also been utilizing AI-driven tools like facial recognition, social media monitoring, deportation targeting, and vehicle tracking. You can read more about our immigration surveillance FOIAs here.
As part of our investigations, we are seeking access to public records regarding the administration’s use of AI. To date, we have sent dozens of records requests to more than fifteen government agencies asking for the following types of documents: communications between agency employees, agency validation studies and privacy impact assessments; records documenting the creation of particular AI systems; results of tests of particular AI systems; agency guidance and directives; source code, training materials, or parameters given to AI systems; searches, instructions, queries, or inputs into AI systems; and outputs, recommendations, or reports created by AI systems.
What We Received
The Department of Veterans Affairs released records confirming its use of AI to repeal agency guidance documents that are not sufficiently aligned with the far-right Trump-Vance agenda. Read more about those records here.
We have also received a GSA policy review guide titled “Policy Review Playbook: A guide for agency policy review to ensure alignment to Trump Administration priorities” that confirms the Trump administration has directed federal agencies to rapidly employ AI to carry out its deregulatory agenda. Created by GSA in February 2025, the guide calls on agencies to integrate automation to remove or revise “outdated” rules, orders, directives, and regulations, including to identify those that may have been impacted by Trump administration executive orders. It also calls on agencies to “utilize AI services to review all new administration documents and categorize them based on the documents’ content and the administration’s priorities,” as well as use website scraping and “AI-assisted policy impact analysis and recommendation.”
From a HUD production, we discovered that Scott Langmack forwarded an email thread called “HUD Dereg Spreadsheets” in late May and early June 2025, which included a fully unredacted attachment describing in detail how to carry out the administration’s “deregulation goal.” Langmack writes, “The spreadsheet you have received represents the system’s thorough analysis conducted on the regulatory sections relevant to your program group’s operations; this analysis has identified specific sections/clauses for complete or partial elimination based on established criteria.” Previous reporting indicates HUD AI tools may include SweetREX or the DOGE AI Deregulation Decision Tool, which Langmack may be referring to here as “the system.” Based on the parameters of the search for which this is a responsive record, we can infer that the spreadsheets generated as the basis of this process were created through a process involving AI. The rest of the document spells out very clearly the administration’s step-by-step process for implementing the deregulatory agenda.
We are waiting for a meaningful response to the majority of our FOIA requests from the government.
