The most common answer attorneys get to the attorney client confidentiality AI question comes from the AI itself, and that is exactly why so many firms have not moved. Ask a general chatbot whether it is safe to use with client files and you get a careful, hedged, technically accurate paragraph that resolves nothing. No lawyer acts on a paragraph like that. So they do not act at all.

Four months, one vague answer, zero progress

A solo practitioner told me how that plays out. She went to an AI seminar in April, bought a subscription the same day, walked out motivated, and asked the tool one question about client confidentiality before she did anything else.

I am really concerned about my client confidentiality. So basically I asked it, and it gave me some answer. And I was like, I am almost too freaked out to really get involved. So I never really got any further.

Four months. Paid subscription running the whole time. And she was not wrong to be careful. Her clients, she said, are really coming to us for this high level confidentiality, and that is one of the things she totally assures them of. That is a real promise made to real people, and she was not going to gamble it on what a chatbot wrote about itself. Her caution was correct. Her source was not.

I have watched this same freeze in consultation after consultation this summer. Every attorney client confidentiality AI conversation follows the same arc. Most writing on AI ethics for lawyers stays up at the level of principle, which is useless to someone deciding on a Tuesday whether a financial affidavit can go into a prompt. The answer is not a principle. It is a setup.

Attorney client confidentiality AI comes down to three concrete artifacts

When a firm hires us, we do not hand over a philosophy of data. We put three things in place, and none of them are judgment calls. The first is the account tier. A team or business plan rather than a consumer one, because the tier is what determines how your data gets handled. That is a purchasing decision, made once, and it is the most consequential choice in the process.

The second is language in the engagement agreement. Here is how I put it on those calls. You need a team plan. It requires us to put some language in your retainer agreement that you are using firm-owned and operated AI systems that operate on your firm’s own internal database. That retainer AI disclosure is not a warning label. It is an accurate description of how the firm works, written where the client reads it before signing rather than discovered later.

The third is internal and it is the one attorneys skip. You also have to have some forms in your internal firm management files that you produce to all of your staff that talks about AI safety use. A written law firm AI policy, sitting in the management file, distributed to every person who touches a matter, so the standard belongs to the firm instead of to each individual’s private guess about what is fine. A paralegal improvising a rule at seven at night is not a policy.

There is a fourth item that is less an artifact than a standing rule. Not every model tier is appropriate for client data, and knowing which tier is which belongs in the setup rather than in someone’s memory three weeks later. We make that assignment during installation and write it down. Systems with real rules also decline to do things, which is the behavior you want out of them. I wrote about a build that refused to draft a petition and was right to refuse.

Those three artifacts plus that rule are what LAL sets up for a firm, and together they settle the attorney client confidentiality AI question in practice. I am describing our installation practice, not telling you what any authority requires of you.

You already answered this question about SharePoint

Here is where attorney client confidentiality AI stops feeling exotic, and it is the part that unfreezes people.

It basically is the equivalent of you using SharePoint or OneDrive in terms of who can have access to it. It’s not overly complicated.

Every firm storing a financial affidavit in Microsoft 365, a mediation summary in OneDrive, or a full matter file in SharePoint already made a decision about a vendor holding client data on servers the firm does not own. You read the terms, you picked the business tier, and you moved on with your life. The AI question is that same question asked about a newer category of vendor, and the analysis you already ran applies. What changes with firm-owned systems is that the work runs against your own material, your own templates, your own saved language, rather than against a general pool of somebody else’s.

None of this is overly complicated. It is just time consuming. That distinction matters, because time consuming has a solution and complicated feels like it does not. An afternoon of setup, done once, in the right order, ends a four-month freeze.

Sometimes the fear is other lawyers, not the data

One solo would not describe her own operations to me at all until she had a lawyer-to-lawyer assurance first. Can we have confidentiality if I tell you stuff about the business, she asked, that you will not tell any other lawyers on the street about my issues. Because everybody wants to kill everybody. I agreed flatly, and then I told her the truth, which is that what she was about to describe was completely ordinary.

That objection is not about data security. It is about shame. The attorney worried that a peer will learn her billing goes out three weeks late, that her file naming is chaos, that she still has a legal pad system nobody else can read. Naming that honestly does more good than another paragraph about encryption. Nobody’s back office looks the way they would want it photographed. The firms that get somewhere with this are the ones that say the messy part out loud early, which is a different problem from the one where there is no room to learn anything while you are putting out fires.

The recording objection is real, and it costs you automation

Some attorneys will not record or transcribe anything, and I do not argue with them. There are some lawyers that are just very uncomfortable with transcribing things, and if that is the case, it can still be done without transcription, but there is a level of automation you just cannot attain.

Here is the tradeoff in practical terms. If a consultation, a strategy session, or a hearing debrief is never captured, then a human being types the summary, builds the timeline, and pulls the follow-up tasks out of memory. That work still happens. It happens slower, and it happens at the end of a day that was already long, which is the same structural problem as keeping the firm’s knowledge in your own handwriting. A firm can decline transcription and still run good systems. It just runs them with a person doing the intake step, and the person is the constraint.

If you are somewhere in month two of your own freeze, book a free exploration call and bring the confidentiality question with you. We will walk the account tier, the retainer language, and the staff policy in one sitting, and you will leave knowing what has to be true before a client file goes anywhere near a model.

Nothing here is legal advice, and each attorney remains responsible for their own compliance with their bar’s rules.