Shadow AI in law firms: staff are already using it, so what next?
By the ChamberOne team 5 minute read
Somewhere in your firm, someone is using an AI tool you have not approved. It may be a free chatbot on a personal phone, a browser extension that rewrites emails, or a transcription app on a laptop. They are not being reckless. They are trying to finish a job. The usual response is to ban the practice, and the usual result is that it goes underground. This article offers a different approach: find out, offer something better and set rules people can live with.
Key points
- Staff use unapproved tools to get work done, so treat it as a sign of an unmet need.
- Bans push use out of sight and discourage people from reporting mistakes.
- Find out through conversation and an anonymous questionnaire, not by punishing honesty.
- Sort uses by risk, and prioritise anything with client confidential material or unverified output.
- Offer an approved tool quickly, with a one-page rule and a no-blame way to report mistakes.
Why people do it
Staff reach for unapproved tools for ordinary reasons. The approved route is slow, or does not exist. A deadline is close. The free tool is easy and good enough. Often the person using it believes, reasonably, that they are doing the firm a favour by working faster.
Understanding the motive matters. If you treat it as misconduct you will not hear about it. If you treat it as a signal that the firm has a gap, you will.
Suppose a paralegal in a ten-person firm has been pasting rough notes of client calls into a free chatbot to turn them into file notes, because file notes are always the last thing done on a busy day. She is not hiding it. Nobody ever asked. When the principal finally does, the conversation is easy, and the firm learns that the real gap is a quick way to record calls, which can be fixed without anyone being disciplined.
Why a ban rarely works
A blanket prohibition is easy to write and hard to enforce. Personal phones sit outside your network. Tools appear faster than any block list. Staff who still need the help will find it, and will be less likely to mention a mistake because they are afraid of a reprimand.
A ban also gives the firm a false sense of safety. Partners believe nothing is happening, so nobody checks. The risk grows in the dark.
If you do keep a prohibition on certain things, such as entering named client material into public tools, say clearly what is prohibited and why, and pair it with a route to doing the same job safely. A rule with no alternative is a rule that will be broken.
Find out what is actually being used
Start with conversations, not surveillance. Tell the whole firm that you want to know what tools people use and why, that nobody will be punished for answering honestly, and that the aim is to provide something better.
- Ask each team what repetitive work they would most like help with.
- Run a short, anonymous questionnaire on which tools are in use and for what.
- Ask your IT provider what it can see, such as browser extensions and app installs on firm devices.
- Look at approved software for features that have quietly added AI.
- Ask suppliers, such as transcription services and e-signature tools, whether AI is in use on your data.
Sort uses by risk
Not every use carries the same exposure. Using a chatbot to suggest a better word for a general email is different from pasting a client's medical report into it. Sort what you find.
- Low risk: no client information, no reliance on output, such as rewording non-confidential text.
- Medium risk: client-related work in which no confidential detail is entered, with a lawyer reviewing the result.
- High risk: client confidential or privileged material in an unapproved tool, or unverified output used in advice or court documents.
Offer an approved alternative
The strongest control is a good approved tool that is as easy to use as the unapproved one. Choose it after asking the data questions: what is stored, whether input is used for training, who can see it, where it is processed and who the sub-processors are. Record the answers.
Make it available quickly. If approval takes six months, the shadow tool will be firmly established by then.
Consider approving in stages. A first approval might cover non-confidential uses only, such as rewording general text, while the data questions for client material are still being answered. Staff then have something sanctioned straight away, and the firm keeps its tightest rules for its most sensitive work.
Review what people ask for once the approved tool is live. Requests it cannot meet, such as transcription or translation, point to the next gap to fill and keep people from drifting back to the old tools.
- Easy to write and hard to enforce.
- Personal phones sit outside your network.
- Staff are less likely to mention what they use.
- Gives a false sense of safety.
- The strongest control.
- As easy to use as the unapproved tool.
- Made available quickly.
- Can be approved in stages.
Write rules that fit on one page
People follow rules they can remember. Keep the policy to a page and put it where staff will see it.
- Which tools are approved, and for what uses.
- What must never go into any tool, such as certain categories of client information.
- The checking required before output is used, especially citations.
- How to record that AI was used on a matter.
- Who to ask, and how to report a mistake without blame.
Keep watching, kindly
Review the position every few months. New tools and new features appear constantly. Ask again what people are using. Keep the report-a-mistake route open and respond to it constructively, because it is the best early warning you have.
A simple test of whether the culture is working: when did someone last tell you about a mistake involving a tool? If the answer is never, the likeliest explanation is not that none occurred.
Check the guidance from your regulator or bar on supervision and on your responsibility for work done with technology. A firm remains accountable for work whichever tool was used.
Include contractors, locums and temporary staff in the rules. They often arrive with their own habits and tools and are the group least likely to have been told what the firm expects.
- 1Ask again what people are using
- 2Respond to reported mistakes constructively
- 3Look at requests the approved tool cannot meet
- 4Fill the next gap
- 5Review again in a few months
- Then back to step 1.
How ChamberOne handles this
- The firm chooses the model for each AI feature, GPT-5 or Claude, or switches AI features off for the firm.
- Each AI output stays a draft until someone accepts it.
- Every AI call is recorded in the audit log with the model and token counts.
- The Security and AI page says what is sent to the AI provider and on what terms.
What it does not do
- With AI features on, the text needed for a request is sent to the AI provider for that feature, outside the country where ChamberOne is hosted. With them off, nothing is sent.
- ChamberOne does not control tools staff use elsewhere, and it does not replace a firm policy.
- AI output can be wrong or incomplete and does not replace a lawyer's review.
This article is general information about the law, legal practice and technology, current when it was published. It is not legal advice. Laws and guidance change: check the sources linked above and take advice on your own circumstances from a lawyer qualified where you are.
Read next
Sri Lanka's Anti-Corruption Act 2023: a guide for practitioners
What the Anti-Corruption Act No. 9 of 2023 provides, how CIABOC exercises its powers, and what advisers must understand about offences, declarations, and a proposed 2026 amendment.
6 minute readClient confidentiality and AI tools: a checklist before you paste anything in
The questions to answer about retention, training use, access, hosting and sub-processors before client information goes into any AI tool.
5 minute readWriting a law firm AI policy: an outline you can adapt
Six sections that a short, usable AI policy needs, with the questions each one should answer.
4 minute read