Family law firms hold some of the most sensitive information any business sees: bank statements, tax returns, medical records, children’s school records, messages between spouses and, sometimes, evidence of abuse. That makes many family lawyers cautious about AI, and the caution is reasonable.
But caution doesn’t have to mean avoidance. The firms that use AI safely aren’t the ones that ignore the risk. They’re the ones that ask the right questions before any client data goes in. This guide covers those questions, what a sensible firm policy looks like, and how to talk to clients about it.
This is practical guidance, not legal advice. Check your own state bar’s opinions on generative AI, which can go further than the ABA’s.
What the rules say
The ABA’s Formal Opinion 512, issued in July 2024, didn’t create new rules for AI. It applied the existing ones: competence, confidentiality, communication, supervision and reasonable fees.
For confidentiality, the practical takeaway is simple. Before putting client information into an AI tool, a lawyer should understand how that tool handles the information, who can see it and whether it could be disclosed. In some situations, the opinion contemplates getting the client’s informed consent.
In other words, the question isn’t “Can we use AI?” It’s “Do we understand this specific tool well enough to use it with this client’s data?”
Why family law raises the stakes
Every practice area handles confidential information. Family law has a few features that raise the stakes further.
Third-party data. A divorce file is full of information about the other spouse, the children and sometimes extended family, none of whom are your client.
Children’s information. School, medical and counseling records for minors deserve extra care.
Safety concerns. Addresses, schedules and communications can matter for clients dealing with domestic violence.
Sealed and protected records. Some filings and records are restricted by court order or statute.
Emotional volatility. A data incident in a contested custody case can do real harm, quickly.
That’s not a reason to keep using email attachments and shared drives as the default. Those carry their own risks. It’s a reason to choose tools deliberately.
The vendor questions that matter
Before any client data goes into an AI tool, get written answers to these questions.
Is the platform independently audited? Ask for the audit type and a copy of the report or a summary under NDA.
Is client data used to train models? The answer you want is no, in the contract, not just in a marketing FAQ.
Who can access each matter? Inside your firm, inside the vendor, and any outside parties you invite.
Can access be limited and time-bound? Co-counsel, experts and clients should see only what they need, for as long as they need it.
Where does the data go? Which subprocessors and AI model providers receive it, and under what terms.
What happens when we leave? How you export your data and how the vendor deletes it.
Does the tool show its sources? A tool that cites the document behind each answer is easier to supervise, which is part of the duty too.
For reference, Deliberately.ai is SOC 2 compliant, never uses client data for model training, controls access by matter and lets firms grant third-party access that expires automatically. Hold every vendor, including us, to the same questions.
A firm AI policy in one page
You don’t need a 40-page policy. You need one page your team will actually read.
Approved tools. List the AI tools approved for client data. Everything else is approved only for non-confidential work.
What never goes into an unapproved tool. Client names, financial records, children’s information, medical records and anything under a protective order.
Review before release. Every AI-produced document is reviewed by a person before it reaches a client, a court or opposing counsel.
Verification. Facts and citations are checked against the source before they’re relied on.
Client communication. Engagement letters explain, in plain language, that the firm uses approved technology to organize and analyze case information.
Training. New staff learn the policy in their first week, and it’s revisited when tools change.
Billing. Time is billed for time actually spent. Efficiency gains shouldn’t turn into inflated hours.
Talking to clients about AI
Most clients don’t object to technology. They object to surprises.
A short, honest explanation goes a long way: “We use secure, approved software to organize your documents and help us find information quickly. Your information is not used to train AI models, and an attorney reviews everything before it goes anywhere.”
If a client has specific concerns, such as safety concerns in a domestic violence matter, discuss them directly and adjust access accordingly.
Safer by design, not just by policy
Policies matter, but design matters more. The safest workflows are the ones where people don’t have to make a risky choice to get their work done.
When the AI tool already has the matter file, nobody copies sensitive excerpts into a public chatbot to get an answer. When access is limited by matter and expires automatically, nobody forgets to revoke an expert’s access. When every answer cites its source, supervision is quick enough that it actually happens.
A checklist before you start
Written answers to the seven vendor questions
A one-page AI policy, shared with the whole team
Engagement letter language about technology use
A review step for every AI-produced document
Your state bar’s latest AI guidance on file
The bottom line
Family law firms don’t have to choose between AI and confidentiality. They have to choose tools built for confidential work, set a short policy and keep a person in the review loop.
If you’re evaluating AI for your firm, our page on AI for law firms covers how we approach security and governance, and you can book a demo to ask us the vendor questions directly.
Related reading: Confidentiality in the digital age: protecting your privacy during divorce
