Clients, confidentiality and AI: the conversation firms keep avoiding

There is a conversation happening in almost every law firm about AI and client confidentiality, and it is happening in the wrong room. It takes place in risk committees, in IT security reviews, in long email threads between the general counsel’s office and the innovation team. What it almost never involves is the client. Firms have decided, mostly without deciding, that confidentiality is something to be resolved internally and then quietly applied - and that the less said to clients about AI, the better. I think that instinct is understandable, and wrong, and getting more expensive by the month.

Silence is a policy too

When a firm doesn’t talk to clients about AI, it hasn’t avoided taking a position. It has adopted one of two, usually without admitting it. The first is quiet use: the tools are switched on, fee-earners use them, and everyone hopes the question never comes up. The second is quiet prohibition: the tools are banned or so restricted that nobody touches them, and everyone hopes the client never asks why the work takes as long as it did five years ago.

Both are bets on the client not asking. And clients are asking. Panel reviews, outside counsel guidelines, security questionnaires - the AI questions are arriving in writing, often drafted by an in-house team that is already using these tools itself. A firm that has stayed silent then has to answer cold, under procurement pressure, in a document that will be read side by side with its competitors’ answers. That is the worst possible setting for a conversation that needed nuance.

What clients are actually worried about

When you do sit down with a general counsel and talk about this, the worry is rarely the one firms imagine. Very few clients think that a machine reading their documents is itself a breach of confidence. They’re worried about much more specific, and much more answerable, things. Where does the data go? Is it kept, and for how long? Is it used to improve someone else’s model? Who at the vendor can see it? If something goes wrong, will you be able to tell me what happened and to which matters?

Every one of those questions has an answer, and a firm that has done its diligence properly already knows the answers. The problem is not that the answers are bad. It’s that nobody has written them down in language a client can read, because nobody expected to have to.

Confidentiality was never a sealed room

It helps to be honest about how confidentiality actually works in a modern firm, because the AI debate often proceeds as if client information currently sits in a locked drawer. It doesn’t. It lives in cloud email, in hosted document management, with e-disclosure providers, transcription services, outsourced print rooms and counsel’s chambers. Lawyers have been sharing confidential information with third-party systems for decades, and the professional answer has always been the same: know who you’re sharing it with, on what terms, and for what purpose.

AI doesn’t change that principle. It changes the vocabulary, and it adds a few genuinely new questions - retention, model training, sub-processors - that deserve proper answers. But it is a vendor diligence question, not a philosophical one. Framing it as the latter is how firms end up paralysed, treating something they already know how to do as though it were unprecedented.

A clause is not a conversation

The tempting fix is to add a paragraph to the engagement letter and call the matter closed. I understand the appeal. It’s tidy, it’s defensible, and it requires no awkward meetings. But a clause the client signs without reading is consent in the narrowest possible sense, and it does nothing for the relationship. Worse, it can look like exactly what it is: the firm protecting itself rather than informing the client.

The clause has its place. It just shouldn’t be the first time the client hears about any of this. If the first mention of AI is in the small print, the client will reasonably wonder what else is in the small print.

How to actually have it

The firms I’ve seen handle this well do something unglamorous. They write a single page, in plain English, that explains which tools they use, for what kinds of work, where the data goes, and what they will never do with it. Then they put it in front of clients before they are asked, ideally in a conversation rather than an email.

The best versions go further and offer a choice. Some clients will want AI used aggressively on their matters because they want the cost and speed benefits. Some will want it restricted on particular matters for reasons that are entirely their own. A firm that can say “here’s what we do by default, here’s how you can change it, and here’s what changing it means for time and cost” has turned a risk conversation into a service conversation. That is a very different meeting.

And the best firms say what they don’t know. If a question about a vendor’s retention terms hasn’t been fully answered yet, saying so builds more credibility than a confident answer that later turns out to be wrong. Many general counsel were private practice lawyers themselves. They can tell the difference.

The firms that wait for the client to raise AI will spend the next few years answering questionnaires defensively, one panel review at a time. The firms that raise it first get to frame it - as care, as transparency, as a firm that understood the question before it was put to them. Confidentiality has always been about trust, not secrecy. The conversation firms keep avoiding is simply the latest place to earn it.