The emergence of the “Department of Government Efficiency” (DOGE) within the Department of Housing and Urban Development (HUD) recently sparked controversy, as internal efforts to modernize governance took a turn into the opaque world of artificial intelligence. Reports indicate that DOGE members, including individuals like recent university graduate Christopher Sweet and tech entrepreneur Scott Langmack, were tasked with auditing agency regulations. Their primary mandate was to leverage AI tools to identify government rules for potential rescission or cancellation, essentially using algorithms to streamline—or dismantle—existing bureaucratic frameworks. While the administration framed this as an innovative efficiency drive, the transition from human-led policy debate to automated decision-making has left many career HUD employees frustrated and concerned about the accuracy of these AI-generated conclusions.
The lack of transparency following this initiative has now moved from a workplace grievance to a legal battle. Democracy Forward, a nonprofit legal organization, filed a Freedom of Information Act (FOIA) request to better understand how exactly these AI tools were influencing policy. The response they received from the federal government was largely one of non-compliance. HUD withheld more than 100 documents, relying on obscure and arguably questionable legal justifications, including a so-called “AI privilege” that does not exist in standard law. They also attempted to invoke presidential communication privileges, a protection usually reserved for the highest levels of the executive branch and their closest advisors, to shield internal notes on how these software systems were programmed to evaluate public housing rules.
Peering into the few document titles that were made public reveals just how integrated AI had become in the agency’s internal deliberation process. Files such as “GPT defined Econ Analysis approach” and “RegulatoryAnalysisPrompt.pdf” suggest that Langmack and his team were actively feeding prompts into large language models to conduct regulatory analysis—the very process that dictates how government programs impact millions of Americans. By labeling these as “deliberative AI input,” the agency effectively tucked them away from public scrutiny. This raises a fundamental question: if the government is using inscrutable AI tools to justify cutting programs or changing regulations, how can the public ever confirm that the decisions are based on objective data rather than a software error?
Experts in technology and ethics are sounding the alarm, noting that the “black box” nature of these tools is a recipe for disaster in public policy. Tori Noble, an attorney from the Electronic Frontier Foundation, points out that AI is notorious for “hallucinating” facts, reflecting deep-seated societal biases, or simply misinterpreting the complex legal nuances of federal regulations. When a human writes a policy, we can track their logic, debate their values, and hold them accountable in court. When a prompt is run through an AI, the trail of reasoning becomes obscured, making it impossible for citizens to challenge the logic behind a decision that might directly affect their homes, their neighborhoods, or their access to social services.
The deeper issue at play here is the current void in federal legislation regarding AI accountability. Unlike other sectors where transparency is legally mandated, there are currently no laws in the United States requiring agencies to disclose when AI has been used to draft or alter public policy. This “wild west” approach to governance means that as artificial intelligence is quietly integrated into everyday agency functions, the public—and even mid-level civil servants—are being left in the dark. Without a requirement for a “disclosure statement” or a clear audit trail, the government is essentially outsourcing its moral and legal responsibility to machines that cannot be held answerable for the real-world consequences of their recommendations.
Ultimately, the goal of government is to maintain the trust of its citizens, a goal that cannot be achieved through secrecy or technical evasion. Mark Fagan, a lecturer at the Harvard Kennedy School, suggests that if AI is going to remain part of the government’s toolkit, a protocol of transparency is not merely “nice to have”—it is essential for survival. By being open about how these tools are being used, the government could build confidence rather than suspicion. However, the current strategy of withholding documents and creating novel pseudo-legal privileges effectively does the opposite, fueling the belief that automated efficiency is being prioritized over democratic process, accountability, and the basic right of the public to understand how their country is being governed.
