Clean Slate, Dirty Data – An Audit of Algorithmic Automated Criminal Expungement Laws

From the abstract:

“Automatic criminal record restriction laws, which suppress records at the initiative of the state, rather than the individual, present a promising new way to provide second chances to the one in three American adults that live with a criminal record. Buoyed by the success of drug legalization measures and the “Clean Slate” movement, the number of states required to automatically suppress eligible criminal records has expanded dramatically, from a few pilot efforts to around half of all states. With the new laws comes a redemptive role for algorithms in the criminal justice system. Yet, to date, there has been limited empirical analyses of the extent to which algorithmic and automated Clean Slate laws can actually deliver the relief they promise. This Article offers an empirical assessment of criminal records restriction laws, analyzing thousands of commercial background checks and government records for their “fidelity” to the law. Within the sample of studied background checks, it finds that only a small share (2% on average among the studied states) included records restricted by law, a rate far lower than comparator states lacking similar restrictions. Reductions in the prevalence of restricted records correlated with rule changes, suggesting a clearance rate of around two-thirds or more of eligible records. This suggests that record restrictions can succeed where petition-based processes have failed, and offer a way to achieve mass records relief as a counterweight to mass criminalization. But fidelity to automatic criminal record restriction laws was far from perfect, raising the risk of ambiguity rather than equity in clearances. As algorithmic and automated measures become more widespread, so will the need for robust governance mechanisms. This Article calls for measures that shift not only the burden of implementation but oversight to the state, through government audits and a requirement to make administrative comment, complaint, and correction processes available. These and related measures would go far towards realizing the delivery, not just promise, of redemption through automation.”