Why prints, evidence and testimony stay with an officer
This job sits in two places at once. One half is at the scene: dusting surfaces for latent prints, lifting and preserving them, photographing and diagramming the layout, then bagging, labeling and logging what gets collected. The other half is the file room: classifying and comparing prints, keeping criminal history and incident records, and releasing or withholding information under state law. Both halves are judgment work with a signature attached.
Software is good at reading, sorting and searching. It is not good at deciding which doorframe is worth processing, how to lift a print off a curved surface without destroying it, or what to do when a scene has been disturbed before anyone arrived. Those calls happen once, in real conditions, and they cannot be re-run later.
Then there is court. An identification officer may have to explain, under cross-examination, how a sample was collected, who touched it and why a comparison was called a match. Chain of custody is the product here. A model can suggest candidates; a person has to own the conclusion. That accountability is the main reason the work has not moved.
What AI does, what it helps with, what stays human
The automated slice is the paperwork end: pulling records from databases on request, formatting and checking reports, flagging missing fields, redacting standard identifiers in disclosure packets. That share of task time reads as 0% in the task split above this narrative.
The assisted slice is bigger in practice. Automated fingerprint systems have returned ranked candidate lists for decades, and newer tools sharpen photographs, stitch scene images together and cluster similar documents in a records backlog. An officer still verifies each hit. That assisted share prints as 21%.
What is left to people is the core: processing a scene, handling and securing physical evidence, making and defending a comparison decision, and judging what a statute allows to be released. The share of task time the task list puts in that group is 79%, and our coverage score, which asks how much of the work AI can handle today, sits at 13 out of 100. You can read how that number is built on the coverage method page.
The evidence, and what is still missing
There is no direct, published test of AI against people in this job. Our quality-parity evidence grade is D, which means the question has not been measured, so we publish no parity number for it. Algorithmic fingerprint matching has been studied for years as a retrieval tool, but retrieval is not the same as a sworn identification.
Two things would settle it. First, a blind comparison study: certified latent print examiners and an AI system judging the same casework prints, with error rates and inconclusive calls reported for both. Second, an audit of records disclosure decisions, checking whether an automated reviewer withholds and releases the same material a trained officer would under the same statute. Until work like that exists, treat any confident claim in either direction with care. The quality-parity method explains what each grade requires.
When this could change
Most likely after 2042 (8 in 10 of our scenarios). The chart above this section shows that window, and the replacement-year method explains what it measures and what it does not.
Two things could pull it earlier. Cheap scene capture is the first: handheld 3D scanners and body-worn video that produce a complete, timestamped record with less manual photography and sketching. The second is price. Software licensing for records and document work runs far below a staffed post, and the median pay for this occupational group is $93,790 a year (BLS, 2025), so agencies under budget pressure will automate the filing side first.
Two things hold it back. Rules are one: evidence handling, retention schedules and public records law are written around identified, accountable people, and courts expect a witness. Hardware is the other. The robotics panel above places the physical part of this job in the dexterous humanoid tier, which is the hardware that does not exist as a reliable product yet. Public-sector hiring is also steady rather than shrinking, with employment projected to change by 0.2% between 2025 and 2035 (BLS, 2025) across roughly 114,430 jobs.
What to do: get certified in the parts of the job that carry a signature, because those are the tasks the file room cannot hand over.
How to stay needed in identification and records work
Lean into the three tasks that stay with people. Scene processing, where you decide what to collect and how. Evidence custody, where you are the documented link between a scene and a courtroom. And comparison decisions you can explain out loud, step by step, to a jury or a defense attorney.
Two skills raise your floor. One is testimony: plain explanation of method, limits and uncertainty, including how an automated candidate list was used and verified. The other is records law and data governance, which is becoming the skill agencies lack as AI tools start touching retention, redaction and disclosure. If you can audit what a tool did to a file, you are the person who signs off on it.
Nearby jobs are worth a look if you want to move sideways. Detectives and criminal investigators share the casework side, police and sheriff’s patrol officers share the scene side, and court, municipal and license clerks share the statutory records side. You can also see where this job sits among law enforcement occupations, read the wider government sector picture, or put two roles beside each other on the compare page. Our full scoring method is open, and the jobs that mostly need a person appear on our list of the safest jobs from AI.