Why public defenders can't rely on prosecution-first tools

Public defenders deal with an overwhelming amount of discovery every day. The state sends hours of body-worn camera footage, dashcam video, and audio files.

When you are drowning in discovery, a review tool is a necessity.

But how do you pick the right one for your office? Features and technology play a part. But security, client data, and AI ethics matter more in the decision-making.

In this article, I'll walk through the risks you should weigh before adopting a tool for your office.

The convenience trap

You get hundreds of gigabytes of discovery for a single case. Pulling raw video down from a state portal takes hours.

That is why Axon's Evidence.com and NiCE Justice's Evidencentral look appealing. The footage already sits on servers the state uses. You skip the slow process of downloading, converting, and re-uploading large files. It removes a real administrative burden.

But the company that builds the platform also builds the body cameras and Tasers officers wear. So the system is built around prosecution workflows first. NiCE Justice shares that same model.

When you review discovery inside a system built for the state, the line starts to blur.

One tool can't serve both sides.

Jeff Sherr puts the same idea as a hiring question. Which paralegal would you hire?

The hidden activity record

Reviewing evidence is strategic work, and it is where a defense takes shape.

When you use a prosecution-first system, every time you mark a timecode, cut a clip, or write a note inside a shared portal, you create a log in a system that belongs to the other side.

Think about what that log shows: the exact order you work through things, the clip you played nine times, the file you merely opened once.

That is your theory of the case, written down as logs in someone else's software.

The conflict you inherit

A conflict exists when your duty to someone else could limit your work for a client. A vendor serving both sides can do exactly that.

When a funding body chooses a tool for you, you inherit the vendor relationship without necessarily having a say in how that vendor handles your work. So it is worth asking: what gets shared, who can see it, and who ultimately controls the system?

The same AI and support staff may handle work for both sides. Your files, feature requests, and activity logs can shape software built primarily for the other side. Vendor staff may see your file names and activity logs without ever opening a piece of evidence.

And if the vendor's biggest client is your opposing side, you depend on that vendor for access to your own work. Your service, pricing, and support are ultimately controlled by a company whose interests may not be aligned with yours.

How can you avoid the risk that comes with AI tools?

1. Watch out for these terms

When you sign up for software, you are also agreeing to the vendor's terms. The wording matters. Look closely at what those terms allow the vendor to do with your data.

1. "Use such Machine Learning for testing, tuning, optimizing, validating, or otherwise enhancing the analytics, models, or algorithms underlying the System." Under terms like these, your client's data could become material used to improve a product that may be sold to anyone, including the prosecution.
2. "To law enforcement authorities as we reasonably feel is necessary." That gives the vendor discretion to decide when disclosure is necessary, rather than requiring a formal legal process.
3. "Unless Vendor determines [it is necessary to disclose Client Content], in its sole and absolute discretion." That puts the decision in the vendor's hands, rather than requiring your consent or a legal order.
4. "All changes are effective immediately when we post them." A vendor could change its AI or data-use rules overnight. You cannot warn a client about a risk you don't know about.

2. Ask these questions before signing the contract

Send them in writing. Keep the answers. Do not sign until you have clear answers to all of them.

1. Are our uploads, transcripts, notes, and strategy privileged work product under the contract? Who owns them?
2. Will you confirm our uploads and AI outputs never train any model, yours or a third party's?
3. Do you sell to police, sheriffs, prosecutors, or investigators? And what keeps the staff serving them away from our data?
4. If you get a subpoena, will you refuse unless legally forced, and tell us right away so we can seek relief?
5. Would police reading our content or usage data, without a warrant, count as a breach under our contract?
6. Name every subprocessor that can touch our data. Will you tell us before that list changes?
7. On exit, can we export everything in a usable format, and will you certify deletion?

That is the short version. Jeff and Maya's guide has the full questionnaire.

It also scores vendors across seven risk areas and includes a contract addendum you can attach to any deal. Download it free →

Where we stand

To handle the ever-increasing volume of discovery, you have to make use of AI. But the real problem lies in doing it ethically.

At Reduct, we don't serve the prosecution. We've written that commitment into our policy. We're built for the defense, and we work directly with defense teams to shape the product around what is most useful to you.

If you are weighing a discovery review tool, we'd love to jump on a call and share a bit more about us.