Privilege-Safe AI for Law Firms

Attorney-client privilege protects confidential communications made for the purpose of legal advice. Nothing about that definition excludes software. What matters is whether a particular deployment keeps the communication confidential, and whether the firm can show it.

Where the risk actually sits

Not in the phrase artificial intelligence. In specifics: who else can read the content, whether it is retained, whether it trains a shared model, and whether a third party could be compelled to produce it. Those are answerable questions, and the answers vary enormously between products.

What privilege-safe requires in practice

• Firm content processed under the firm’s control, not pooled with other tenants.
• No training on client content, stated contractually rather than in marketing copy.
• A BAA where protected health information is in scope, covering sub-processors.
• Access logging, so the firm can show who saw what and when.
• Retention and deletion the firm sets, not the vendor.

Confidentiality is not the only duty

Competence and supervision apply too. An attorney who files work they did not verify has a supervision problem regardless of what produced the draft. This is why a review gate matters as much as an encryption claim: it is the mechanism by which professional responsibility is actually discharged.

Consumer tools are a different question

Pasting a client's medical record into a general-purpose chatbot is a materially different act from processing it in a system your firm contracted for, with terms covering retention and training. Most firm policies should say so explicitly, because in the absence of a policy people will use what is convenient.

Get it in writing

If a vendor cannot put its data handling in the contract, the assurance is worth what it is written on. Ask for the terms, read the sub-processor list, and keep the version you were sold.