AI finds thousands of outdated laws hidden in policy sludge
When Bureaucracy Traps Itself in Its Own Rules
Imagine a city bylaw from 1975 that still requires paper forms to be filed in triplicate for a permit that no longer exists. This is not a hypothetical. For decades, cities and states have accumulated thousands of such outdated provisions. They sit in the bylaws, unrepealed and unenforced, yet still legally binding. Civil servants must navigate this “policy sludge” every day.
The term describes obsolete processes and stale reporting requirements that do nothing but trap workers in red tape. A simple request can trigger a cascade of pointless steps. A program meant to help citizens can become dysfunctional because the rules governing it were written for a different era. The sludge does not just slow things down — it actively prevents good policy from working as intended.
The problem has a specific shape. It is not a single bad law. It is a mountain of small, forgotten ones. Each one, on its own, seems harmless. Together, they form a barrier that takes years to dismantle through traditional legislative processes. Repealing a single outdated bylaw can require a committee hearing, a public comment period, and a full council vote. Multiply that by thousands, and the task becomes impossible.
An Experiment in Measuring the Unmeasurable
Researchers from the University of Pennsylvania and the University of Chicago set out to see if artificial intelligence could do what human bureaucrats cannot. The study, led by Professor Cary Coglianese of Penn’s Carey Law School and Professor Rebecca Willett of the University of Chicago, tested whether large language models could identify obsolete regulations in state and municipal codes They designed an experiment. The AI would scan entire bodies of municipal and state law. Its task: identify every provision that had become obsolete, contradictory, or simply irrelevant.
The results were dramatic. The AI found thousands of provisions that human reviewers had missed. In one state alone, it identified over 12,000 sections of code that were either duplicative or no longer served any purpose These were not minor typos. They were entire regulatory frameworks that had been superseded by newer laws but never formally removed.
The comparison with earlier manual efforts is stark. Previous attempts to clean up state codes relied on volunteer committees of lawyers and legislative staff. They worked for years and still only managed to review a fraction of the total text. The AI completed a comparable review in weeks. It did not get tired. It did not overlook a clause because it was buried in a subparagraph. It found the sludge that human eyes had learned to ignore.
The Detail That Shows How Far We Still Are
The AI’s discovery has immediate practical consequences. The researchers identified a specific case: Medicaid reporting requirements. In one state, providers were required to submit a monthly report on a service that had not been offered since 2003. The report took an average of four hours to complete. It served no purpose. The AI found it, and the state was able to eliminate the requirement.
But the application still faces a hard limit. The AI can find the sludge, but it cannot remove it. That still requires a human decision. A legislator must vote to repeal. A governor must sign. A civil servant must update the forms. The AI provides the map, but the journey remains manual.
Coglianese and Willett acknowledge this gap They describe the AI as a tool for triage, not a solution in itself. The real work — the political work of actually cutting through decades of accumulated red tape — remains ahead. The sludge is visible now. The next step is to decide what to do with it. As Coglianese put it, ‘AI can show us the mess, but only democratic processes can clean it up.