DOGE has used AI in housing policy. The government won’t say how

Members of the so-called Department of Government Efficiency (DOGE) who were working in Department of Housing and Urban Development (HUD) Using artificial intelligence to inform policy decisions. Now, the agency appears to be denying FOIA requests for information about the development and use of AI tools, and the way policy decisions are made, according to documents obtained at a FOIA request by Democracy Forward, a nonprofit legal organization.

last year, WIRED reported Christopher Sweet, then a third-year student at the University of Chicago, joined the DOGE team at HUD, along with Scott Langmackwho came to DOGE from a real estate technology startup called Kukun. Sweet’s primary focus, according to HUD employees who spoke to WIRED at the time, was using artificial intelligence to determine the agency’s rules for potentially canceling, or canceling, contracts. Part of a similar effort Through the government.

At the time, HUD staff told WIRED that employees had been reached out to provide feedback on regulations that had been flagged by the AI ​​for repeal. But other employees described the effort as redundant.

Sweet graduated from the University of Chicago in June with a degree in economics. Langmack now serves as executive director of AI editing at the Office of Management and Budget (OMB), the Executive Office of the President, according to his LinkedIn.

More than 100 documents requested by Democracy Forward regarding HUD’s use of artificial intelligence in decision-making have been redacted. Among the reasons HUD cited for not releasing the documents, the non-existent artificial intelligence privilege and the presidential communications privilege were real but generally applied only to the president and his direct advisors. Several of the withheld documents, whose names were shared in the FOIA but whose contents are unknown, appeared to indicate that HUD’s DOGE team was using artificial intelligence tools to help make policy decisions.

Sweet, Langmack, HUD, OMB and the White House all responded to requests for comment.

One document, titled “GPT-Specific Economic Analysis Approach 11 10 25.docx,” which was owned by Langmack, was exempt from the Freedom of Information Act because it was classified as “deliberative AI input.” Another document titled “RegulatoryAnalogyPrompt.pdf,” also belonging to Langmack, appeared to indicate that the DOGE team was looking into creating claims to conduct regulatory analysis. Several other documents redacted for being part of the deliberation process are labeled as a form of “regulatory analysis” of various HUD programs, though it is not clear whether artificial intelligence was used in their creation.

The lack of transparency around how AI tools are used to create or change policy is particularly troubling, because the tools are known hallucination, Show biasOr just plain Getting things wrong. “We don’t always have to know how to use tools,” she says. “So having access to claims is really the best way to be able to see what officials are using these tools for and how much harm those uses might be causing.”

There are currently no laws in the United States that require the government to disclose whether AI has been used to create rules, policies, or regulation.

“If AI is being used for policy evaluation as one of the tools in the toolkit, then I think at this point in the development and use of AI, it’s a good protocol to point to that,” says Mark Fagan, a lecturer at Harvard’s Kennedy School. “Partly to try to build confidence in the use of AI in government.”