The Department of Government Efficiency — DOGE, the entity created by Elon Musk to cut waste and automate processes — has used artificial intelligence to steer housing policy. But when a journalist requested the records under public access laws, the Department of Housing and Urban Development (HUD) walled them off, partly by invoking a legal privilege that simply doesn't exist.

The incident is more than an administrative oddity. It goes to the heart of how democracies plan to embed algorithms into decisions that affect citizens' lives. An automated decision on subsidies, housing eligibility thresholds, or priority allocation can shift resources and fates, and without verification, the line between efficiency and arbitrariness collapses.

HUD's opacity is a structural signal: when a public agency claims a nonexistent privilege to avoid accountability, it means that AI oversight is seen as a nuisance, not a responsibility. For those working in on-premise and self-hosted deployment ecosystems, this is a warning bell. The dominant cloud model, concentrated in few hands, makes it harder for citizens and watchdog bodies to trace decision chains, verify model versions, or reconstruct training data. That's the opposite of what's needed in government contexts, where digital sovereignty should be a prerequisite.

We don't know whether DOGE used an LLM, a statistical model, or a more complex pipeline. But the lack of transparency makes the specific technique irrelevant: the point is that a decision-making infrastructure was steered without leaving a public trail. Those hurt first are the most vulnerable — people who depend on housing assistance — and the trust required for any digital modernization of public administration is eroded.

The episode comes against a backdrop where many governments push for rapid AI adoption, often without building the independent audit tools that make the process verifiable. The result is a paradox: automation accelerates while understanding of how it works is locked down. For companies and institutions evaluating deployment architectures, the lesson is clear: without explainability mechanisms and immutable logs, AI in the public sector risks becoming a black box that amplifies inequalities instead of reducing them.

HUD chose silence. DOGE chose speed. Citizens, for now, have no choice.