Software bugs reportedly keep Arizona inmates jailed past release dates

A failure to update critical prison management software has kept hundreds of incarcerated people in Arizona behind bars longer than they should be, according to a whistleblower report. Employees for the Arizona Department of Corrections have known about the bug since 2019, Phoenix-based NPR affiliate KJZZ reported. The flaw follows a change to state law that the software simply cannot handle and has not been updated to deal with. Arizona has one of the highest imprisonment rates in the country, with drug possession convictions being one of the highest drivers behind the numbers. An amendment (PDF) to Arizona state law in June 2019 created a mechanism through which inmates convicted of certain nonviolent drug offenses can earn credits toward early release. Eligible inmates who complete a program such as a GED equivalent or substance abuse treatment while imprisoned can earn three days' credit for every seven days served and shorten the length of time they spend behind bars to 70 percent of their assigned sentence. The Arizona Department of Corrections' own FAQ (PDF) about the program specifies that neither inmates nor their families should contact anyone to request an eligibility review. "This is done automatically based on system programming, which generates a list for Time Comp," the document reads. "The reviews are being completed based on the projected earliest release dates." According to the whistleblowers, however, that system cannot identify eligible participants and still does not make those calculations. "We knew from day one this wasn’t going to work," a source in the Department of Corrections told KJZZ. "When they approved that bill, we looked at it and said 'Oh, shit.'" The whistleblowers say they began making "repeated internal warnings" to the department's IT officials in 2019. KJZZ obtained a copy of an October 2020 bug report detailing how the software, called ACIS, fails to keep up with the law. "Currently, this calculation is not in ACIS at all," the report reads. "ACIS can calculate 1 earned credit for every 6 days served, but this is a new calculation." The department confirmed to KJZZ that it is aware of the problem with ACIS. "Data is being calculated manually and then entered into the system," spokesperson Bill Lamoreaux told KJZZ, saying that the department has identified at least 733 inmates who are eligible to take part in the early release program but are not yet enrolled.

Jail (mis)management software

There's an entire sprawling US industry of "corrections" software out there. On one level, jail and prisoner management systems make sense: you need to know who's in a facility, where they are, what their medical needs are, who their legal representation is, and so on. As elements of the criminal justice system are increasingly handed over to algorithms for management, however, real people are becoming more affected by the flaws in these software packages. The ACIS deployment in Arizona took three years longer than it was supposed to and consistently went over budget, local media reported at the time. Since it was finally implemented over Thanksgiving weekend in 2019, ACIS has remained plagued by bugs, sources told KJZZ. Several modules in the software have failed to perform correctly, "including modules that track inmate health care, head counts, inmate property, commissary and financial accounts, religious affiliation, security classification, and gang affiliations." And when human error introduces something incorrect into an inmate's file, employees may not have a way to remove or correct the mistake. "In one instance, there was a disciplinary action erroneously entered on an inmate's record,” a source told KJZZ. "But there's no way to back it out. So that guy was punished and he wasn’t able to make a phone call for 30 days. Those are the kinds of things that eat at you every day." Software suites used by police departments and courts—long before someone is subject to prison management software—have also been widely criticized in recent years, both for a lack of transparency and for exhibiting racial bias. ProPublica in 2016 published a deep, detailed report finding that courts' risk assessment software ranked Black suspects as more likely to commit further crimes than white suspects who allegedly committed similar acts. Researchers in 2018 analyzed the software and found that groups of random amateurs had equal accuracy when projecting suspects' re-offense rates; both people and algorithms were wrong about a third of the time.