DOGE Used AI to Shape Housing Policy. Now HUD Is Blocking Requests to Learn How.

Key points

• The Department of Housing and Urban Development (HUD) withheld more than 100 documents in response to a public-records request about its use of AI in policymaking.
• A nonprofit called Democracy Forward filed the Freedom of Information Act (FOIA) request, a formal legal tool citizens and organisations use to obtain government records.
• HUD cited a so-called "AI privilege" to block some records, an exemption that legal experts say has no basis in federal law.
• One withheld document, belonging to a DOGE team member, was titled "GPT defined Econ Analysis approach 11 10 25.docx" and labelled "deliberative AI input."
• No US law currently requires the government to disclose when AI has been used to create or change rules and regulations.

A small team inside the Department of Housing and Urban Development spent months using AI tools to flag federal rules for potential cancellation. The public still does not know exactly how those tools worked, or how much weight their output carried. HUD is refusing to say.

The nonprofit legal group Democracy Forward filed a Freedom of Information Act (FOIA) request, the standard legal mechanism citizens use to demand government records. HUD responded by blocking more than 100 documents. The reasons it gave ranged from the genuinely debatable to the legally invented.

The team doing this work came from the so-called Department of Government Efficiency (DOGE), the White House-linked initiative that sent staff into federal agencies early in 2025. As Wired first reported, that team at HUD included Christopher Sweet, then a third-year economics student at the University of Chicago, and Scott Langmack, a former executive at a property-technology startup. Sweet's main job, according to HUD employees at the time, was using AI to identify agency rules that could be scrapped or revised. Langmack now serves as executive director of deregulation AI at the Office of Management and Budget.

Among the withheld documents, two stand out. A file called "GPT defined Econ Analysis approach 11 10 25.docx," belonging to Langmack, was blocked under the label "deliberative AI input." A second file, "RegulatoryAnalysisPrompt.pdf," also Langmack's, appears to show the team was building prompts, the text instructions that tell an AI chatbot what to do, to carry out regulatory analysis.

Did HUD have the legal right to withhold these records?

In most cases, no, according to legal experts. The government has nine recognised exemptions under FOIA. HUD leaned heavily on Exemption 5, which protects the back-and-forth between human government employees while they are working through a policy decision. The idea is to protect candid internal debate. But John Davisson, deputy director of enforcement at the Electronic Privacy Information Center, says that exemption does not stretch to conversations between a person and an AI chatbot. "AI systems, computers are not entitled to candor," he said. HUD also cited a "presidential communications privilege" for at least one prompt document, which raised further questions among legal observers about where those instructions originated.

For ordinary people, the practical concern is straightforward. AI tools, including the large language models that power chatbots like ChatGPT, are known to produce confident-sounding errors, reflect built-in biases, or simply get facts wrong. If such a tool helped identify which housing rules to cancel, and the public cannot see the prompts or outputs, there is no way to check the work.

"Having access to the prompts is really the best way to be able to tell what officials are using these tools for and how harmful those uses might be," said Tori Noble, a staff attorney at the Electronic Frontier Foundation.

No legislation currently forces any US government agency to label a policy decision as AI-assisted. Until that changes, the public may simply have to take the government's word for how these tools are being used.