Washington Governor Jay Inslee says a programming glitch has caused as many as 3,200 inmates to be released early since 2002, by a median amount of 49 days
Joseph O'Sullivan / The Seattle Times :
Context & Ripple Effects
Washington Governor Jay Inslee has disclosed that the state's offender-release calculation software has been miscomputing sentences for over a decade, letting up to 3,200 inmates out a median 49 days early since 2002. The disclosure lands in a documented pattern of corrections software failing at exactly the arithmetic its users depend on: whistleblowers later described how Arizona's inmate management system has been unable to interpret sentencing laws since 2019, leaving hundreds eligible for release still incarcerated.
What makes this more than an IT story is where the field is heading. While Washington's bug erred by releasing people early, [[a:884581|Louisiana's 2024 parole law hands the parole board's power to TIGER, an algorithm issuing immutable risk scores]] that blocks thousands from early release — showing states automating sentencing outcomes in both directions without proven audit practices in either.
First-order effects
- Inslee must now decide whether to return the affected offenders to custody or accept the early releases — a decision that directly affects crime victims' families, the roughly 3,200 ex-inmates whose lives have moved on, and the Department of Corrections staff who ran the flawed calculations for thirteen years before anyone caught it.
Second-order effects
- Other states running similar sentencing-calculation systems face immediate pressure to audit their own code, since Arizona's parallel failure shows the defect class is systemic to off-the-shelf offender-management software rather than a one-state coding slip.
- Vendors and internal IT teams behind corrections software can expect heightened procurement scrutiny and demands for independent verification of sentencing logic, raising costs and timelines for states modernizing these systems.
Third-order effects
- Corrections is becoming an algorithm-governed domain on both ends of the sentence — buggy release calculators on one side, immutable risk scores like TIGER on the other — pushing toward a structural demand for audited, transparent sentencing software as a condition of ceding human discretion to code.
The trend: State criminal-justice agencies are automating sentencing and release decisions faster than they are building the audit capacity to catch when that automation errs.