Confidentiality · November 10, 2026

Shadow AI: how to inventory every tool that already touches client files

Most firms that say they don't use AI yet already do. It's on a phone, in a browser extension, or switched on inside software the firm pays for every month. Before you write an AI policy, find out what's already in use. Here's a plain way to do that.

The tools are already in the building

Ask the partners at most small firms whether the firm uses AI and you'll often hear "not yet." Ask the staff and you'll get a different answer. Someone summarizes deposition transcripts with a chatbot on their phone. Someone else turned on the meeting notetaker that came free with the video call software. The e-signature service added a "summarize this agreement" button last spring, and the practice management system now drafts emails.

None of this is sinister. People reach for whatever makes a long day shorter. But it means the first job for a firm is finding out where AI already is, what it's touching, and who said yes to it. People call this shadow AI: tools in use that nobody at the firm ever reviewed.

Why it matters more than the usual software sprawl

A firm has always had a few apps it didn't know about. A note-taking tool here, a PDF converter there. AI tools are different in one important way: their whole job is to read what you give them. A document converter handles a file and hands it back. An AI tool reads the contents, may keep the text, may keep your prompts, and may send both to companies you've never heard of to do the work.

So every unreviewed AI tool is a small, quiet decision about client confidentiality that nobody actually made. Bar ethics guidance keeps coming back to the same expectations: understand the technology you use, protect client information, and supervise the people and tools doing work on a matter. You can't do any of that for a tool you don't know exists.

Start with a list, not a policy

The instinct is to write a rule first. Resist it for a week. A policy written before you know what's in use either bans tools people quietly keep using, or blesses tools nobody has looked at. An honest inventory comes first, and it's less work than it sounds.

Here's a routine a small firm can run in a few afternoons.

What to write down for each tool

Keep it to one row per tool in a plain spreadsheet. Fancy software for this is overkill at a small firm. For each one, capture:

That third column tends to be the most revealing. The same tool can mean very different things depending on whether a staff member signed up with a personal email or the firm holds the account and can see the settings. We've written about how to read an AI vendor's security page, which helps fill in the fourth.

Nobody gets in trouble for what turns up on the list. The inventory turns decisions nobody made into decisions somebody made, on purpose, with the facts in front of them.

Sorting what you found

Once the list exists, most tools fall into one of three groups.

Fine as is. Tools that never see client information, or firm accounts whose terms someone has actually read and accepted. Write down who approved them and move on.

Fixable. A useful tool on a personal account, or a built-in feature with a setting that should be changed. Move it to a firm account, adjust the settings, and tell people what's allowed in it.

Stop. Anything where client documents are going into a tool whose terms nobody can find or explain. Stop it, and then do the part that keeps the inventory honest next time: ask what problem the person was solving. Someone was pasting transcripts into a chatbot because summarizing them by hand takes hours. If the firm takes the tool away and offers nothing, the habit goes back underground.

Keep it current

An inventory is a snapshot. New features arrive in software updates, staff change, and someone will find a new app next month. Put a short review on the calendar every quarter, and add one line to onboarding: before you use an AI tool on firm work, tell the person responsible for the list. Make that easy and people will do it.

Where we fit

A lot of shadow AI exists because the work is real and the approved options are thin. One answer is a tool the firm actually controls. The Tiber River Legal Workbench runs on hardware the firm owns and reads documents where they already sit, so there's one place client files get reviewed instead of a dozen browser tabs. It produces source-linked review aids, keeps a record of what it touched, and leaves every decision with the attorney. If you're weighing that trade, our post on owning your AI instead of renting it covers the upkeep honestly.

Tell us what turned up on your list

The Tiber River Legal Workbench is a matter intelligence appliance for litigation and personal injury practices, running on hardware your firm owns. The tools your staff reached for on their own say a lot about where the work actually hurts. We're inviting a small number of firms to shape it as design partners.

Start the conversation

Or write to josh@tiberriverconsulting.com

← All posts