The Justice Sector Is Stuck in Pilot Phase.

Here’s Why….

Across courts, regulators, and justice agencies, artificial intelligence is everywhere, and nowhere.

Justice institutions are clearly experimenting at scale. In the AI-adoption tracker, there are 131 recorded adoption instances, with 85 classified as experimental and only 8 described as operational (with 37 still unclear or unknown). That gap is the story: activity is high, but institutional commitment remains cautious.

This isn’t because the technology is incapable. It’s because justice systems run on institutional legitimacy.

When decisions affect liberty, rights, and public trust, deploying AI is not simply an IT decision. Institutions must be able to demonstrate that AI use is accountable, transparent, and consistent with procedural fairness. Without that assurance, scaling AI becomes reputationally and legally risky, even when the tools themselves work.

And the legitimacy burden is rising quickly.

The Justice Sector AI-Policy Frameworks tracker now tracks 99 frameworks, with 25 already in force. The dominant themes are consistent: accountability (62) and transparency (58) lead the list, followed by human oversight (40), fairness/bias (36), privacy/data protection (35), security (33), and audit/assurance (26). Governments are codifying expectations for oversight faster than institutions are building the operational machinery to meet them.

This is where many pilots stall.

Moving from experimentation to institutionalisation requires more than model performance. Organisations need governance infrastructure: clear lines of responsibility, auditable records of where AI is used, visibility into how outputs influence decisions, and defined pathways for review and contestability when something goes wrong.

Procurement patterns reveal where institutions feel comfortable deploying AI today.

In the Courts AI-procurement tracker there are 24 recorded AI-procurements, with 16 already implemented. But the intended uses are telling. Transcription (16) and search (7) dominate, both functions support human work. Case management appears only marginally (3), and triage or intake barely registers (1). Institutions appear far more willing to deploy AI where it assists decision-makers than where it might be perceived as shaping outcomes.

And when AI does begin to shape outcomes, controversy follows.

The AI-related controversies database now tracks 89 matters, including 52 strong-evidence cases. The most common issues are accountability (67) and transparency (51), followed by procedural fairness (32), accuracy/reliability (30), bias/discrimination (29), and privacy/data protection (22).

These are legitimacy failures, the points at which justice institutions face the sharpest scrutiny.

Across the datasets, the pattern is consistent: experimentation without institutionalisation.

The short term outlook for AI adoption in justice systems is not determined primarily by new AI capabilities. The decisive factor will be governance infrastructure - the practical systems that allow institutions to demonstrate responsible use in ways that can survive audit, oversight, litigation, and public challenge.

Until justice systems can show not only what AI does, but how it is governed, many pilots will remain exactly that: pilots.

The challenge ahead is not technological. It is institutional.

And solving it will be critical to Justice Sector modernisation.