The first question before a court uses AI is not, “Which tool should we buy?” It is, “What work are we changing?”
That question is at the center of Tennessee's current court-modernization work. The Administrative Office of the Courts submitted its fourth statewide technology report on September 14. The report says the AOC has held more than 100 business-process workshops with up to 300 court clerks, judicial officers, and staff. It reports that 1,765 court processes and their corresponding process flows have been reviewed and documented.
That is not an AI deployment announcement. It is more useful than that. It is a record of the work that must be understood before technology changes it.
The report says the AOC is negotiating with a selected vendor for statewide case management and e-filing. It identifies a future statewide AI strategy as one initiative on its Action-Oriented Roadmap, with goals and guiding principles aligned with judicial objectives and ethical standards. Tennessee has not adopted that strategy, selected an AI vendor, or announced AI use in the courts. The report instead shows a practical sequence: understand the process, identify the people and data involved, and build the controls before implementation.
A current U.K. court record offers a concrete comparison. HM Courts & Tribunals Service has published an algorithmic-transparency record for BenchNotes, a production tool used by Immigration and Asylum Chamber judges to transcribe oral decisions in real time. HMCTS says the speech models remain within its secure infrastructure, are not trained on personal data, and produce output that users must always review and validate. It also says the tool does not make decisions and that judicial training is provided.
The point is not that Tennessee should copy BenchNotes. The point is that a court can describe an AI-enabled workflow in enough detail to answer the public's basic questions: What does it do? What information does it use? Who sees it? Who checks it? What remains human?
Four questions before AI changes a court process
- What exact work is changing? A court should begin with a defined task, not a broad promise of efficiency. Is the work scheduling, document organization, transcription, public information, research support, or something else? The answer determines the risk.
- What information enters the tool? The data boundary should be clear before someone copies information into a system. Case facts, confidential material, sealed information, and personal data require a different analysis from public forms or a published notice.
- Who reviews the output before it matters? A transcript, summary, search result, or draft may save time. It should not become the court's work product until a responsible person checks it. The review point must be built into the process, not added after an error.
- What record remains? A court should be able to explain the tool's purpose, the approved users, the data rules, the review step, and the correction path. A process map makes that explanation possible before a problem forces it.
Courts do not need to settle every future AI question before they improve one present process. But they should know the work well enough to tell when a useful assistant has begun to change responsibility, confidentiality, or the record.
Question for local discussion
Before a proposed AI tool reaches a court workflow, can the court show what work it will change, what data it will receive, who will review its output, and what record will remain?