Is AI safe for client data?

Here's the uncomfortable reality at most firms: while leadership debates whether AI is safe, associates and staff are already using it. They're pasting a client's facts into a free chatbot to draft a demand letter or summarize a transcript, because it saves an hour. The efficiency is real. So is the exposure, and that gap between unofficial use and official policy is exactly where confidentiality problems start.

So let's answer the question directly. Is AI safe for confidential client data? It can be, when it's set up correctly. It is not when client information is dropped into consumer tools that may retain it or train on it. The difference between those two scenarios is configuration, documentation, and supervision, not the technology itself.

What your duty actually requires

Start with Model Rule 1.6. It requires a lawyer to make reasonable efforts to prevent the unauthorized disclosure of, or access to, information relating to the representation of a client. It does not say "never use new technology." It says be reasonable about protecting client information. The question is therefore not whether AI is allowed, but whether your use of it meets that reasonableness standard.

Then there's ABA Formal Opinion 512, issued July 29, 2024, the American Bar Association's first formal ethics guidance on generative AI. It confirms that lawyers may use these tools, and it lays out the duties that come with them: competence (understand the tool's benefits and risks), confidentiality, supervision of the work product, candor to the court, reasonable fees and billing, and, importantly, a duty to evaluate the AI vendors you rely on. Opinion 512 is the closest thing the profession has to a blueprint, and your own state's Rules of Professional Conduct sit on top of it.

The real risk: consumer chatbots, used casually

When people ask whether "AI" is safe, they're usually picturing the wrong threat. The danger isn't a properly configured system. It's the free, public chatbot that nobody at the firm decided to use but everyone is using anyway. Some consumer tools retain inputs and may use them to improve their models, which is the opposite of what confidentiality requires. Combine that with no firm policy, no record of what's being used, and the occasional fabricated citation, and you have a genuine problem, one that already has its own genre of news stories.

The fix is not to ban AI and fall behind. It's to replace the unofficial, risky version with an official, safe one.

What "safe" actually looks like

A defensible AI setup in a law firm has a handful of non-negotiable features:

• Zero data retention. The tools are configured so that what you put in isn't stored or used to train any model. This is the single most important control, and it's available on the right platforms.
• A written vendor agreement. The terms with your AI provider are documented, so the data handling isn't a guess, it's in writing. This is exactly the vendor vetting Opinion 512 calls for.
• A firm AI use policy. A clear, written policy covering what may and may not be used with AI, how use is disclosed, and a decision tree for when not to use it at all.
• Human supervision, always. AI drafts; a lawyer reviews and is responsible for every output. Nothing client-facing or court-facing goes out unreviewed.
• Grounded drafting. Document workflows built on your own templates and materials, with verification steps, so fabricated cases and invented facts don't survive to a filing.
• Documentation for audit. If the bar ever asks how your firm uses AI, you have the answer in writing before the question comes.

Notice that none of those are about being clever with technology. They're about discipline, the same kind of discipline you already apply to client files and trust accounts.

A practical checklist before you use any AI tool

• Does this tool offer zero data retention, and is it turned on?
• Do I have a written record of how the vendor handles our data?
• Does my firm have an AI use policy that staff have actually read?
• Is a lawyer reviewing every output before it leaves the firm?
• Is there a clear list of what we will never put into AI?
• Could I explain our AI use to a client, or to the bar, without flinching?

If you can't check all six, the answer to "is this safe?" is "not yet."

Where we fit, and where we don't

This is the work we do: building and documenting AI systems for firms so they can answer every one of those questions with a confident yes. Because the systems are configured for zero retention, backed by vendor agreements, and paired with a written policy and staff training, the safe version becomes the easy version, which is the only way official policy ever actually beats unofficial habit. You can read the full approach on our compliant AI for law firms page, or our guide on how to write an AI use policy for your firm.

One important boundary: we're an AI implementation vendor, not your law firm. We don't give legal or ethics advice, and we don't opine on your specific obligations. Our founder's law degree informs how we build and document systems so your attorneys can satisfy the duties they already have. Your state's rules govern, and your lawyers remain responsible for every output. That distinction isn't a disclaimer for its own sake, it's part of doing this correctly.