A youth-led courtroom agenda for transparency, contestability, and accountability as automated systems shape liberty, evidence, surveillance, and punishment.
Automated systems are no longer only questions for engineers, vendors, or AI policy summits. They are becoming courtroom questions.
A risk score can shape whether a young person is released or detained. A facial recognition lead can start a prosecution. A school surveillance record can follow a student into discipline, probation, or court. An AI-generated image, transcript, report, or message can enter a file before anyone knows how to challenge it.
This agenda starts where broad AI principles usually stop: the courtroom. Before automated material shapes detention, prosecution, adjudication, supervision, or juvenile outcomes, the rules have to be clear enough for people to see it, challenge it, and know who is responsible.
The question is not whether a decision is made by a human or a machine. The question is whether a system improves upon existing practice while remaining transparent, contestable, and accountable.
We build on the work of youth organizers, civil rights groups, researchers, lawyers, and technologists who have already warned about algorithmic harm. Our focus is narrower: courts, juvenile proceedings, schools, probation, detention, and policing, the places where technical claims become state power.
That is why these calls are addressed to the actors who can change the rules: courts, lawmakers, agencies, and companies.
Two kinds of automated systems are already shaping American criminal proceedings, and the law is behind on both. The first is algorithmic decision tools, including risk scores and predictive instruments, most of them not AI at all, that already inform bail, sentencing, and supervision decisions. The second is artificial intelligence, consumer chatbots, generative systems, and AI-altered media that are newer to the courtroom and far less settled. This agenda addresses both, and it names them separately, because they raise distinct problems that demand distinct answers.
The rules governing the first have lagged for a decade. In State v. Loomis, a proprietary risk score factored into a criminal sentence even though neither the defendant nor the court could examine how it worked, and trade-secret claims still shield these tools from scrutiny in courtrooms across the country.
The rules governing the second are being written right now, case by case. In February 2026, Judge Jed Rakoff of the Southern District of New York ruled in United States v. Heppner that a defendant’s conversations with a consumer AI chatbot were not protected by attorney-client privilege, one of the first reported federal decisions to reach that question. Within days, major law firms warned their clients that what they type into a chatbot can be discovered and used against them. Law firms warned clients quickly. Ordinary users usually get no court-specific warning at the point of use. Millions now ask AI tools about their legal situations without knowing those words can be obtained by the other side, not because the law of privilege is wrong, but because no one is required to tell them.
The technologies are moving faster than the legal rules governing them, and the people with the least power to contest them are the most exposed.
We are the generation that will live longest under the systems being built right now. Here is what we ask leaders to commit to before an automated system is allowed to shape a person’s liberty.
Risk-assessment tools like COMPAS already inform bail and sentencing, yet defendants are often never told a tool was used, and the companies that build them shield their workings as trade secrets. The same problem reaches forensic and evidentiary software: probabilistic genotyping, facial-recognition systems, AI-assisted transcription, and tools that generate or alter exhibits. In State v. Loomis, the Wisconsin Supreme Court let a proprietary risk score factor into a sentence even though neither the defendant nor the court could examine how it worked. You cannot challenge what you cannot see.
But visibility rules pegged to the case file will miss most of this. Predictive policing software, automated license plate readers, gunshot-detection systems, and social media monitoring tools routinely shape a stop, a search, or a charging decision well before a case exists, and their influence rarely survives into the paperwork. A risk score that helped trigger a stop that ended in arrest will not appear anywhere in that arrest report. That does not mean it was not used. Naming only the tools that already show up in court filings lets every tool operating upstream of them go unaddressed.
When the state uses a risk score to justify detention, supervision, or punishment, that output functions like an accusation. It has to be answerable. Too often, automated outputs arrive with the authority of neutral science even though their assumptions, data, settings, and error rates remain contested or hidden. A defendant can cross-examine a witness. There is no cross-examination of a proprietary instrument whose workings are off limits. Due process requires that evidence or assessments used against a person remain meaningfully open to challenge, regardless of whether they were produced by a human or a machine.
Facial recognition has already produced a string of wrongful arrests of Black Americans, and the foundational research showing these systems fail most often on darker-skinned faces is now a decade old. When a tool causes that kind of harm, responsibility too often vanishes into the space between the agency that deployed it and the vendor that built it.
Children are not small adults, and the labels attached to them follow for life. Predictive and surveillance tools are already entering schools, and data gathered to educate a student can be quietly repurposed to police that same student, a new, automated on-ramp to the justice system built into the institutions young people are required by law to attend.
We are asking scholars, advocates, technologists, and young people who share these commitments to add their names. The technology will keep moving. These principles should not have to be re-argued every time it does.
Names may appear as signatories, advocates, or organizational supporters as submissions are reviewed.
Authored by Nicholas E. Stewart, Justice Education Project