HUD Is Using AI to Cut Housing Rules and Hiding How
HUD is blocking public access to AI prompts used to identify federal regulations for removal, invoking a legal privilege that experts say does not apply to machine-generated outputs.
Documents obtained through a Freedom of Information Act request reveal that operatives from the Department of Government Efficiency working inside the Department of Housing and Urban Development used artificial intelligence to flag federal regulations for removal. HUD is now refusing to release the specific prompts and inputs that shaped those decisions, raising serious questions about accountability in automated policymaking.
The agency is withholding more than 100 documents by citing what it describes as a "deliberative AI input" exemption under FOIA's deliberative process privilege. That privilege exists to protect candid internal debate among government officials. Nonprofit Democracy Forward and legal experts say it does not extend to machine-generated outputs. A computer, they argue, has no deliberative capacity that the law is designed to protect.
The AI work inside HUD was led by Christopher Sweet and Scott Langmack, who reportedly used large language models to scan and flag rules for potential rescission. Document titles surfaced in the FOIA request include "GPT defined Econ Analysis approach" and "RegulatoryAnalysisPrompt," suggesting the agency ran structured prompts through AI tools to inform deregulation decisions. Critics say that without access to those prompts, the public cannot determine whether the AI was pointed toward a predetermined outcome or whether it produced errors that were acted upon without review.
The concern goes beyond legal process. AI models are known to hallucinate, misread context, and reflect the biases embedded in their training data. When those outputs feed directly into housing policy, the consequences fall on real people, including low-income renters and public housing residents who have little recourse if flawed analysis drives rule changes. The United States currently has no law requiring agencies to disclose AI involvement in rulemaking, leaving a significant gap in oversight.
What This Means for African AI Governance
The HUD case is an instructive warning for African governments actively building AI policy frameworks. Nigeria's draft National AI Policy and Kenya's data protection infrastructure are both evolving in an environment where pressure to automate public services is growing. The U.S. experience shows that deploying AI in regulatory processes without mandatory disclosure creates accountability blind spots that are difficult to close after the fact. African policymakers have an opportunity to get ahead of this by writing transparency requirements directly into AI governance rules, specifying that any AI tool used to inform or draft public policy must be documented and its outputs open to scrutiny. Efficiency and accountability are not in conflict, but only if the rules are designed that way from the start.
Calling machine output a deliberative thought is not a legal argument; it is a workaround, and the distinction matters far beyond Washington.
Source: Wired
Written by
StacyAI-assisted news curation. Every story is reviewed by our editors before publication.
