AI for law enforcement
AI for law enforcement, with the officer still the author
One job in this category. Camera audio becomes an AI-assisted first draft the officer reviews, edits and owns.
15-minute fit call. Free 30 to 90 day pilot for up to 50 officers, subject to approval. No credit card required.- Officer reviews, edits and owns
- No facial recognition or predictive policing
- CJIS Security Policy-aligned
- SOC 2 compliant architecture
Author of record
The objections that stop these programs start with authorship
Where AI report drafting has been paused or restricted, the argument has been the same each time: an officer attests under oath to a narrative, so the narrative has to be the officer's. That is the design question this page is actually about.
How authorship works in CLIPr
The officer reviews, edits and owns the final report.
- The officer is the author of record. CLIPr produces a draft. The officer reads it against the recording, corrects it, and is the person who signs and submits.
- The output is an AI-assisted first draft. It is never a finished or filed report, and it is never described as one. Nothing leaves CLIPr as a completed document.
- The tool assists, it does not decide. AI does not supply observations, sensory detail, training-based inference, legal judgment, probable cause or credibility findings. Those stay with the officer.
- The recording is the source, and the recording has limits. A camera captures what was said and what was in frame. It does not capture a smell, a read on a room, or an inference drawn from training. The officer supplies those, and any honest description of this technology says so.
- The agency sets the policy. Departments choose which video types are processed automatically, and a department admin trains CLIPr on the agency's own report format.
Settle ownership, retention, deletion, export, subprocessor and model-training terms in the order form and data terms. That is worth doing with every vendor in this category, including this one.
What two states have already written down
California SB 524 was signed on 10 October 2025 and took effect on 1 January 2026. It requires a disclaimer on a report written fully or in part with AI, identification of the AI program used, an officer certification that the report was reviewed and is accurate, retention of the original AI-produced draft for as long as the report is retained, an audit trail, and restrictions on what a contracted vendor may do with agency data.
Utah SB 180 was signed on 25 March 2025 and took effect on 7 May 2025. It requires a disclaimer, an author certification, and a written generative AI policy at every agency in the state.
These are two states, not a national rule, and neither one applies to an agency outside its own jurisdiction. They are still the clearest published description of what a defensible process looks like, which is why they are worth reading before you write a policy anywhere.
Primary sources: California SB 524, Legislative Counsel; Utah SB 180, Utah State Legislature. Statutes change. Confirm the current text and your own state's position with counsel before adopting policy language.
The category, split honestly
Not all AI in policing carries the same risk, and pretending otherwise helps nobody
Most coverage of this subject treats one category. A chief being asked about AI at a council meeting needs the whole map, including the parts CLIPr has nothing to do with.
| AI use case in policing | Where the objection sitsThe argument you will actually face | CLIPrWhere this product sits |
|---|---|---|
| Transcribing recordings | Accuracy in poor audio. Crosstalk and heavily accented speech. Whether the transcript becomes discoverable material in its own right. | In scope. CLIPr returns a timestamped, searchable transcript synced to the recording.CLIPr does not produce certified transcripts. A court transcript vendor is the right choice when a court requires one. |
| Drafting the report narrative | The live one. An officer signs under oath. Three questions follow:
| In scope, and this is the core product.
|
| Redacting for public records | Both directions hurt. Over-redaction breaks transparency. Under-redaction exposes a bystander. Both land on the records division, not on the vendor. | A separate CLIPr service, priced openly. CLIPr publishes 99.5% redaction accuracy with human verification. The human verification is the part that matters. |
| Face matching and identification | The most litigated use in the category. It is the one that produces the wrongful-arrest headlines. | Out of scope. CLIPr's redaction service redacts faces and identities, which is the opposite job. CLIPr does not match a face to a name. |
| Forecasting where or by whom crime will occur | Bias carried in from historical enforcement data, and the community and council resistance that follows it. | Out of scope. CLIPr does not forecast anything. It documents an incident that already happened, from a recording of that incident. |
| Where to start | The deployments that ran into trouble were the contested uses. The ones that did not share a shape:
| That is how a CLIPr pilot is meant to run.
|
Deployment guidance worth reading before you write yours: the COPS Office Dispatch on AI-assisted report writing, which recommends supervisor review against body-camera footage, starting with minor incidents, mandatory officer sign-off and periodic audits.
How it works
Three steps, so the mechanism is not a black box
If you cannot explain a tool to a council member in three sentences, it is hard to defend when someone asks. Here are the three.
A recording goes in, and only the recordings you choose
Officers upload audio or video from the body-worn camera platform the agency already runs. Departments choose which video types are processed automatically, so the scope of the program is an agency decision rather than a default.
There is no dock to install on day one and no camera to swap out.
Scope is a control, not a setting
Restricting the program to a defined set of report types is a recurring recommendation in published deployment guidance, and it is easier to widen a scope later than to narrow one after an incident.
The recording comes back indexed and timestamped
Every clip returns a timestamped, searchable transcript synced to the video, a Topics Index of named topic and subtopic segments, and Question and Answer cards that each carry their own time range.
That indexing is what makes the review step realistic. A supervisor comparing a draft against footage jumps to the marked moment instead of watching the whole recording again.
CLIPr drafts. The officer reviews, edits and owns it.
CLIPr drafts the report and emails the officer when it is ready for review. The officer reads it against the footage, corrects it, and copies the finished report into the agency's records management system (RMS).
A department admin can train CLIPr on the agency's own report format, and a CLIPr dictionary corrects repeated errors so they do not recur on later reports.
The draft is a starting point, not a filing
Output is an AI-assisted first draft, never a finished or filed report, and nothing is submitted without an officer reading it against the source recording.
Before you book anything
Who this is for, and who it is not for
Plenty of people reading about AI in policing are not buying anything, and that is fine. This is for the ones who are being asked to make a decision.
CLIPr fits when
- You have been asked to bring a position on AI to a chief, a council, a prosecutor or a bargaining unit.
- You want to start in the documentation half of the category rather than the contested half.
- You want the officer to stay the author of record on every report that gets filed.
- You would rather measure a result on your own reports than argue about a vendor's number.
Probably not your tool when
- You are evaluating facial recognition, license plate recognition or predictive analytics. CLIPr is in none of those categories.
- Your prosecutor has already declined to accept AI-assisted narratives. Settle that conversation first, because no pilot survives it.
- You want something that files the report without an officer reading it. That is not what this is.
- You are shopping for a records management system (RMS), computer-aided dispatch (CAD), a digital evidence management system (DEMS) or body-camera hardware.
CLIPr states it is actively used by law enforcement agencies across the U.S. Agency size sets the size of the pilot cadre, not whether CLIPr fits.
How to find out for yourself
Run it as an evaluation, not as a rollout
The pilot is the point of this page. It is the only way to answer the accuracy question, the time question and the prosecutor question with your own evidence instead of somebody's slide.
The cadre
Small, willing and supervised
Start with a small group of officers who want to try it, on a defined set of report types, with a supervisor comparing drafts against the footage. A pilot that starts everywhere answers nothing.
The measurement
Your reports, your before and after
CLIPr publishes a reduction in report writing time of up to 50%. Independent research on AI report drafting has not consistently found time savings, so time the work yourself on both sides rather than taking any vendor's number on faith.
The paperwork
Policy in parallel, not afterwards
Utah already requires every agency in the state to have a written generative AI policy. Draft yours during the pilot so it is ready when the decision is.
| Duration | 30 to 90 days. |
| Officers included | Up to 50 officers. The pilot cadre is usually smaller and scaled to the agency. |
| Cost | Free. No credit card required. |
| What you need | Recordings from the cameras you already run, and officers who will use the review step. |
| How it starts | A 15-minute fit call, then pilot setup, subject to approval. |
"This is a game changer. We can finally Protect more and Document less using CLIPr's Generative AI."
Police Officer, Bentonville, ARBring the prosecutor into the pilot, not after it
Bringing your district attorney in before the pilot is easier than explaining it afterwards. Fifteen minutes is enough to scope a pilot you can describe to a prosecutor, a council and a bargaining unit before it starts.
Pilot terms and the report-time figure as published by CLIPr, verified 2026-08-09. Utah's written-policy requirement: Utah SB 180.
Questions command staff ask first
The five that come up on every call
No. CLIPr produces an AI-assisted first draft from the recording. The officer reviews it against the footage, edits it, and owns the report that is filed.
AI does not supply observations, sensory detail, training-based inference, legal judgment, probable cause or credibility findings. If a fact was not in the recording, it has to come from the officer.
Neither. CLIPr documents an incident that already happened, from a recording of that incident. It does not match a face to a name and it does not forecast anything.
The one place CLIPr touches faces at all is its separate redaction service, which detects and blurs them so people cannot be identified in a public records release.
That depends entirely on your state, and anyone who answers it nationally is guessing. California SB 524 and Utah SB 180 are the two clearest published regimes, and both center on disclosure and officer certification rather than on banning the technology.
Read them as a benchmark, not as your obligation, and confirm your own state's position with counsel before you adopt policy language.
Utah is the one state in this research that requires every agency to have a written generative AI policy. Confirm your own state's position and your prosecutor's with counsel. Either way, the pilot is the natural time to draft the policy.
Model policies exist to start from, including Lexipol's generative AI use policy and the Future Policing Institute model policy. Treat any sample language as something for your counsel to adapt, not as text to adopt as written.
Keep the cadre small, limit it to a defined set of report types, have a supervisor compare drafts against the footage, and time the work on both sides of the change so the before-and-after is yours.
Watch the review step itself. If edit rates fall to almost nothing, that is a signal to spot-check rather than a signal that the tool got better.
Find out on your own reports
Fifteen minutes is enough to scope a pilot you can defend to a prosecutor, a council and a bargaining unit. Bring your camera platform, your RMS and your sworn count.
15-minute fit call. Free 30 to 90 day pilot for up to 50 officers, subject to approval. No credit card required.There is a separate self-serve route. If your department has fewer than 50 sworn officers and recently upgraded, or is about to upgrade, its mobile data terminals (MDTs) through a reseller partner, you qualify for a self-serve 30-day trial at app.clipr.ai/agencysignup. Every other agency should use the fit call above.