When COVID-19 swept through correctional facilities in 2020, officials faced an impossible calculus. Prisons and jails were petri dishes for a deadly virus, and the only fast intervention was releasing people. Within months, jurisdictions across the country pushed tens of thousands of incarcerated individuals back into communities far earlier than anyone had planned.

What followed was an unintended natural experiment on a scale that criminal justice researchers rarely get to observe. Thousands of people whom the system had deemed too dangerous or insufficiently reformed to release were suddenly living among us. The outcomes of that experiment challenged some deeply held assumptions about who actually needs to be behind bars.

The data from these emergency releases is now mature enough to examine carefully. What it reveals isn't a simple vindication of mass decarceration or a cautionary tale about reckless leniency. Instead, it offers a more nuanced portrait of how poorly our systems predict individual risk—and how powerfully institutional momentum resists the lessons that disruption teaches.

Multiple studies tracking cohorts released under COVID emergency orders found recidivism rates that were remarkably low—often significantly lower than rates observed for people released through standard processes. A National Institute of Justice review of several jurisdictions showed that the vast majority of people released under emergency provisions were not rearrested for violent offenses within the follow-up periods studied. Some jurisdictions reported rearrest rates for serious crimes in the single digits.

These numbers came with important caveats. The pandemic itself suppressed certain types of criminal opportunity. Courts were backed up, policing patterns shifted, and community conditions were abnormal. Researchers have worked to account for these confounders, and while they moderate the findings somewhat, they don't eliminate the central observation: most people released early did not pose the public safety threat that their continued incarceration implied.

The data also revealed something instructive about the gap between predicted and actual risk. Many of those released had been assessed as moderate or even elevated risk under standard instruments. Their successful reentry suggests that risk assessment tools—already criticized for embedded biases—may systematically overestimate danger for certain populations, particularly older individuals and those convicted of serious but situationally specific offenses.

This doesn't mean that every incarcerated person can be safely released. It means that the threshold we use to justify continued confinement is calibrated far more conservatively than public safety actually requires. The COVID releases didn't prove that prisons are unnecessary. They suggested that prisons are overused—and that the margin of overuse may be considerably wider than even reform advocates had argued.

Takeaway

When a crisis forced mass releases, the predicted public safety catastrophe largely didn't materialize. The gap between how much danger we assume incarcerated people pose and how much they actually pose may be one of the most consequential miscalibrations in criminal justice policy.

Emergency releases were not random. Jurisdictions applied varying selection criteria—age, health vulnerability, remaining sentence length, offense type, and institutional behavior records. Some states focused narrowly on people within months of their release dates. Others cast wider nets that included individuals serving longer sentences for more serious offenses. The variation in these approaches created natural comparison groups that reveal how selection criteria shaped outcomes.

Jurisdictions that applied narrow, conservative criteria—releasing mainly low-level, short-sentence individuals—unsurprisingly saw the best outcome numbers. But the more revealing finding came from places that released broader populations. Even in these jurisdictions, the outcomes were far better than system actors had feared. The categories of people deemed too risky for early consideration often performed nearly as well as those considered safe bets.

This pattern highlights a persistent problem in correctional decision-making: offense category is treated as a reliable proxy for future danger, but the evidence for this is weaker than policy assumes. A person convicted of a violent offense fifteen years ago who has aged, matured, and participated in programming may present far less risk than someone cycling through short sentences for property crimes. Yet the person with the violent conviction label is almost always last in line for any release consideration.

The selection process also exposed the role of political risk aversion. Decision-makers in many jurisdictions chose criteria designed to minimize the chance of a high-profile failure rather than to maximize the number of safe releases. This is rational from an individual career perspective but deeply irrational from a public policy standpoint. It means that people who could safely return to their communities remain incarcerated not because of evidence but because of the political cost structure surrounding release decisions.

Takeaway

Offense labels function more as political shields for decision-makers than as reliable indicators of future risk. When selection criteria were loosened under crisis pressure, the system discovered that many people it had refused to release were never the dangers it claimed them to be.

Perhaps the most sobering aspect of the COVID release experience is how little lasting change it produced. By 2023, prison populations in most states had climbed back toward or even exceeded pre-pandemic levels. The emergency had opened a window, and the system closed it. Understanding why requires looking beyond policy choices to the structural dynamics that drive incarceration numbers.

Part of the rebound was mechanical. Courts that had paused during the pandemic worked through backlogs, feeding new admissions into a system that had temporarily slowed intake. Parole and probation revocations, which had been relaxed during the emergency, returned to pre-pandemic patterns. But the mechanical explanation only goes so far. The deeper driver was that no jurisdiction restructured its sentencing, parole, or charging practices in response to what the emergency data showed.

The institutional actors who control prison population—prosecutors, judges, parole boards, legislators—each have incentive structures that favor maintaining or increasing incarceration. Prosecutors face electoral pressure to be tough. Judges follow sentencing guidelines built on assumptions the COVID data challenged but that no legislature has revised. Parole boards operate under revocation-heavy frameworks that send people back for technical violations. Each actor can point to the others as the source of the problem.

This is what system momentum looks like in criminal justice: not a single decision to maintain mass incarceration, but a distributed network of decisions that individually seem reasonable but collectively reproduce the status quo. The COVID experiment demonstrated that a substantial portion of the incarcerated population could live safely in communities. The system's inability to absorb that finding reveals that evidence alone doesn't change institutions—you have to change the structures that make the evidence irrelevant to the people making daily decisions.

Takeaway

Temporary crises can reveal that a system's assumptions are wrong, but revealing the problem and fixing it are entirely different challenges. Lasting decarceration requires restructuring the distributed incentive systems that reproduce mass incarceration through thousands of individual decisions every day.

The COVID emergency release programs handed criminal justice researchers something rare: real-world evidence on a massive scale about who actually needs to be incarcerated. The answer, by the data's account, is fewer people than we lock up.

But evidence doesn't reform systems on its own. The rapid return to pre-pandemic incarceration levels demonstrates that the forces sustaining mass incarceration are structural, distributed, and largely indifferent to outcome data. Changing this requires more than better research—it requires redesigning the decision architectures that prosecutors, judges, and parole boards operate within.

The pandemic didn't prove prisons are unnecessary. It proved that our threshold for using them is poorly calibrated—and that knowing this changes nothing until we build institutional structures capable of acting on what we've learned.