AI Receptionist
Client Confidentiality When an AI Answers Your Law Firm’s Phones
Every call to a law firm is potentially a privileged communication. The caller describing their situation to your intake staff, or to a machine answering your phones, is sharing information they believe is confidential. That belief doesn’t change because the voice on the line is artificial. If you’re putting an AI receptionist on your firm’s phones, you need to think through confidentiality the way you’d think through it for any new hire, because ethically, that’s what it is: a new member of your intake team.
This guide walks through the confidentiality questions law firms should ask before deploying an AI receptionist, what good answers look like, and the setups that should make you walk away.
The ethical starting point
A lawyer’s duty of confidentiality covers information relating to the representation, and it extends to prospective clients too. When someone calls your firm seeking legal help, what they tell you is protected even if you never take the case. That protection doesn’t depend on who answered the phone. A receptionist, a paralegal, an answering service, and an AI system are all bound by the same principle: the information stays inside the firm.
The practical question is whether your AI vendor’s plumbing honors that principle. Call recordings and transcripts are the obvious risk surface. Ask the vendor directly: are calls recorded, are transcripts generated, where are they stored, who can access them, and can you delete them on demand? If the answers are vague, that’s an answer.
What to require in the vendor setup
Encryption in transit and at rest. This is table stakes. Call audio and transcripts should be encrypted both moving between systems and sitting in storage. Ask which standard they use and don’t accept “we take security seriously” as a substitute for specifics.
Access controls. Your firm’s call data should be visible to your firm and to the minimum vendor staff needed to keep the service running. It should not be used to train the vendor’s models. It should not be accessible to other customers. Ask for this in writing.
Data residency and retention. Know where the data physically lives and how long it’s kept. Set a retention policy that matches your firm’s needs: keep transcripts long enough to be useful for intake review, then delete them on a schedule. Indefinite retention of every intake call is a liability, not an asset.
No training on your calls. Some AI vendors use customer conversations to improve their models. For a law firm, that’s a non-starter. Your callers’ privileged information cannot become training data. Get a contractual commitment, not a verbal assurance.
The intake script itself is a confidentiality tool
How the AI handles the call matters as much as how the vendor handles the data. A well-built legal intake script protects confidentiality by design:
It doesn’t ask for more than it needs. A caller describing a dispute doesn’t need to recite every detail for the AI to book a consultation. Capture the structure of the matter, not the whole story. The details belong in the attorney consultation.
It handles the “am I talking to a lawyer?” question honestly. Callers sometimes assume the person answering works for the firm as legal staff. The AI should identify itself appropriately and never imply it’s an attorney. Misleading a caller about who they’re talking to creates problems far beyond confidentiality.
It knows when to stop and escalate. If a caller starts describing something sensitive and complex, the script should capture the basics and hand off to a human rather than conducting a deep factual interview by phone. Good conflict-check scripting follows the same principle: collect what’s needed to check, then stop.
Conflict checks and the confidentiality overlap
Conflict checking is where confidentiality gets operational. To check a conflict, the system needs the names of the parties involved, which means the caller’s information and the adverse party’s information sit in the same system. Make sure the AI collects conflict-check information separately from the narrative of the matter, stores it in a way your conflict system can use, and doesn’t broadcast party names in notifications or summaries sent to the wrong people.
Also decide up front what happens when a conflict is found. The AI should have a graceful, professional response that doesn’t reveal anything about the existing client or matter. “We’re not able to assist with this matter” is enough. It should never explain why.
What to put in your engagement terms
Consider updating your intake disclosures to reflect the AI’s role. You don’t need legalese on the phone, but your intake process should make clear that calls may be recorded for quality and intake purposes, that information shared is treated as confidential, and how long recordings are kept. Most firms already have language for this from their human intake process. Extend it.
The setups to walk away from
Walk away from any vendor that can’t tell you where your data lives, that trains models on customer calls without a clear opt-out, that stores recordings indefinitely with no deletion option, or that routes your callers’ data through systems you can’t audit. The cost savings of a cheap phone AI evaporate the first time a client asks where their privileged information went and you don’t have an answer.
Confidentiality isn’t a feature you bolt on after deployment. It’s the foundation the whole setup rests on. Get it right, and an AI receptionist is simply a diligent, tireless member of your intake team. Get it wrong, and it’s the most talkative hire you ever made.
I write the agents that run on clinic phone lines and inboxes: the conversation engine and the booking logic behind them, plus the integrations with Pabau, Fresha and Phorest. Everything here comes out of systems we have actually shipped, not a content plan.
