ReportShield
AI-POWERED REPORT REVIEW
Last Updated: September 4, 2026

AI Disclosure

This page explains, in plain terms, what Report Shield's AI does with your report text, what an AI text watermark is, and what that means for you as the officer certifying the report.

What Report Shield does with AI

Report Shield sends the report text you paste in to a third-party AI service. It does two different kinds of work with it.

Analysis. Most of what the app produces is review, not writing: objectivity flags, defense vulnerabilities, legal gaps, charge review, probable cause assessment, timeline, statute and case law lookup, Miranda check. These give you information to act on. They do not produce text for your report.

Rewriting. The Final Report and Polished Supplemental panels are different. Those produce a rewritten version of your narrative in first person, with grammar corrected and language objectified. Your facts are preserved, but the sentences are substantially the AI's construction rather than yours. You review it, approve it, and copy it into your records system.

The rewrite only runs when you open one of those two panels. If you never open them, the AI never drafts narrative text for you.

What a watermark is

Some AI providers now embed an invisible watermark in the text their models generate. It is a statistical pattern in word choice, not a visible mark. You cannot see it by reading the text, it does not change what the text says, and find-and-replace does not remove it. Its purpose is to let the provider confirm later that a piece of text came from their model.

Do Report Shield's outputs carry one?

Possibly, depending on the AI model version in use at the time.

Our AI processing provider applies watermarking to models released on or after August 2, 2026. Models released before that date are not watermarked. Report Shield runs on a pinned model version, and model versions change over time as older ones are retired by the provider.

We are not going to tell you your reports are definitely watermarked, because today that is likely not the case. We are also not going to tell you they never will be, because that is a promise we cannot make about a third party's product roadmap. The honest position is this: assume text the AI writes for you may be watermarked, and decide accordingly.

How much does a watermark actually attach to?

This depends heavily on how much the AI wrote.

Who can check for a watermark?

Access to the provider's detection tool is currently restricted. It is available to organizations with a legal or regulatory obligation to verify AI content: regulators, law enforcement agencies, news media, fact-checking organizations, academic researchers, educational institutions, and certain enterprises with compliance requirements. The provider has said it intends to widen access over time.

Two things worth understanding about that list:

  1. Defense attorneys are not on it today. They may be later. Plan for the possibility, do not panic about the present.
  2. Law enforcement agencies are on it. Another agency, or your own, could in principle check report text. This cuts a direction officers do not usually think about.

What detection can and cannot show

A positive result indicates AI involvement somewhere in that text. That is all.

It cannot show:

A result is a signal, not proof, and it is not conclusive on its own.

What happens to your report data

Separate from watermarking. Report Shield sends report text to our AI processing provider through a commercial API, not a consumer chatbot. Under that provider's current terms, API content is not used to train their models and is retained in their systems for a limited period (currently a maximum of seven days) before deletion. Report Shield does not write your report text to any database of ours. See our Privacy Policy for the full detail, including the named subprocessor.

Why this matters for your reports

You certify your report narrative as your own account of events. If AI-generated text is in that narrative and someone later identifies it as AI-generated, you may be asked how much of the report is yours. That question can come up in discovery, on cross-examination, or in an internal review.

The question is easier to answer when the answer is "all of it."

What we recommend

  1. You can use Report Shield without putting AI text in your report. Run your analysis, work through the objectivity flags, defense vulnerabilities, legal gaps, and charge review, then correct your own narrative yourself. Those panels never write text for you. Only text you copy out of the Final Report or Polished Supplemental panels came from the AI.
  2. If you do use the rewrite, read it line by line before approving. Reject anything that changes a fact rather than the wording. You are certifying it as your account of events.
  3. Read the "Changes Made to Your Report" list shown with every rewrite. It tells you what the AI altered and why. It is the fastest way to see how far the rewrite moved from what you wrote.
  4. Talk to your department's legal counsel or command staff about how AI-assisted report writing fits your agency's report-writing and evidentiary policies. Some agencies have a position on this. Many do not yet.

Questions

Contact support@reportshield.app.