Artificial intelligence is coming to juvenile justice whether agencies are ready for it or not. Vendors are already pitching risk assessment tools, predictive analytics, and automated case management to state directors across the country. Some of these tools will genuinely help. Others will introduce risks that most agency leaders are not equipped to evaluate on their own. Before the next vendor walks through your door with a neural network and a sales pitch, here is a practitioner's guide to the questions worth asking.
The Stakes Are Different In Juvenile Justice
Juvenile justice agencies hold some of the most sensitive data in government. Information about minors, their families, their mental health, their trauma histories. When a private company markets an A.I. tool for adult corrections, the risks are operational. When that same tool gets adapted for juvenile justice, the risks become constitutional. A biased algorithm making detention recommendations for a fourteen-year-old is not a minor technical flaw. It is a decision that can shape the rest of a young person's life.
Improving Agency Workflow: The Best Place to Start with A.I.
Not every application of A.I. in juvenile justice carries the same risk. The tools most likely to deliver real value are operational, not decisional. Automating compliance reporting. Flagging data anomalies in facility records before they become audit findings. Streamlining intake documentation so caseworkers spend less time on paperwork and more time with youth. These applications solve real workflow problems without asking an algorithm to make judgments about a young person's future.
Higher Risk Means Higher Scrutiny
The tools that deserve the most scrutiny are the ones that touch decisions about a youth's liberty. Risk assessment scores that influence detention decisions. Predictive models that flag a young person as high risk before they have done anything to earn that label. These are the applications where agency leaders need to slow down and ask hard questions, because getting it wrong can lead to life-changing consequences.
Questions Most Vendors Are Not Prepared to Answer, But Should Be
Where did the training data come from, and does it reflect the population your agency actually serves? A model trained on adult corrections data or on a demographically different jurisdiction will not perform the way the vendor promises.
What biases does the model carry, and how has the vendor tested for them? Ask for the actual validation studies, not a marketing summary of them.
Who has access to the outputs, and how long is the data retained? A youth's risk score should not outlive its usefulness or end up in the hands of people who should never have seen it.
What happens when the algorithm reaches a conclusion that runs counter to existing decision-making tools? Every vendor should have a clear answer to this: a documented process for a human reviewer to reconcile the discrepancy, and a record of which outcome the agency ultimately relied on and why. If a vendor cannot describe that process, the tool is asking your staff to trust the algorithm over their own validated instruments with no way to check its work.
Has this tool been independently validated in a juvenile justice setting, or only in adult corrections or in a different state? The corrections and juvenile justice market is new enough to A.I. that the evidence base is thin. Do not accept a pitch that treats a thin evidence base as a mature one.
What to Demand Before You Sign
Demand the same rigor from an A.I. vendor that you would demand from any other intervention you adopt. If a therapy program claimed to reduce recidivism without evidence, no agency would adopt it on the vendor's word alone. The same standard should apply to a technology platform.
Demand transparency about how the model works, not just what it outputs. A vendor who cannot explain their model in plain language to your staff has not built a tool your staff can be accountable for using.
Require a plan for human oversight. This concept is known as Human in the Loop. Even if an A.I. platform is used to automate relatively mundane tasks, always ensure that there is a human on your staff who is directly responsible for the A.I.'s outcomes. Some organizations put the A.I. platform on their organizational chart. This means the A.I. reports to a human supervisor who is ultimately responsible for the A.I.'s performance and output.
Never outsource decisions about a youth's confinement status to A.I. This admonition holds whether it is a decision to admit or a decision to release from custody. A.I. can be a powerful tool to aggregate and summarize large amounts of data to assist human decision-makers. But A.I. should never make any final decision that impacts a young person's liberty. That is a critical decision that should only be made by human beings who are supervised in the organizational chain and can be held accountable. When listening to an A.I. vendor's sales pitch, be wary of end-to-end A.I. solutions and the implications at critical decision points.
Demand to know what happens if the tool produces a result or outcome that is inconsistent with your validated instruments or agency policy. Ask how incorrect outputs get recorded and forwarded to the vendor for future retraining, and how often the A.I. is retrained with new data. Write the answers into the contract.
The Bigger Lesson
Artificial intelligence is not inherently good or bad for juvenile justice. It is a tool, and like any tool, its value depends entirely on how carefully it is chosen and how well it is understood by the people responsible for the young people it touches. Agency leaders do not need to become data scientists to evaluate these tools well. They need to ask the questions that most vendors are hoping nobody asks, and they need to demand answers before the contract is signed, not after the first incident.
That is the perspective I bring to agencies evaluating A.I. in juvenile justice. If a vendor is pitching your agency an A.I. platform and you want a second opinion grounded in practitioner experience rather than a sales deck, I can help you ask the right questions before you sign.