← back

Paper Trail

In development · Python

Accountability answers about government AI, straight from the public record.

The problem

Agencies use AI to decide who gets benefits, who is flagged for fraud and who is investigated. The documents that show how those systems are governed are scattered and rarely read, so when a system gets it wrong, it’s hard to tell what accountability, if any, exists. Paper Trail grew out of the op-ed Engineered Silence, which traced that gap across Medicaid, unemployment insurance, tax enforcement and tenant screening.

The solution

Paper Trail answers accountability questions about government AI systems using only the public record. Every answer cites the exact passage it came from. When the record is silent, it says so.

The first version tests the idea on 20 to 30 public government AI contracts, asking does the contract give the government the power to find out the system was wrong, and to do something about it?

Know
Must the vendor document the system’s data, testing and known limits?
Show
Can the agency share that documentation, or its findings, with the public?
Test
Is there a required accuracy level and a right to audit or test?
Act
Can the agency fix, suspend or end the system when it fails?
Leave
Does the government keep its data and the ability to switch vendors?

How it works

Public contractsgovernment AIExtracttext, page by pageAsk5 questionsVerifyevery quoteAnswerwith its sourceNo text layer?Flagged, never read as silentQuote not found?Marked unverified
Dashed lines are safeguards.
Read the README ↗