← All guides

Private AI for Lawyers: Using AI Without Breaching Client Confidentiality

Kiyomi · 7 min read

AI is extraordinary at the work lawyers do most: drafting, summarizing, first-pass research, reworking language. The problem is that the duty of confidentiality does not have an exception for "I was in a hurry," and pasting client facts into a public AI tool can put privilege and your bar standing at risk.

The rule that governs this

Model Rule 1.6 requires you to protect information relating to the representation and to make reasonable efforts to prevent its unauthorized disclosure. Bar associations have been clear that using technology does not suspend that duty. When you paste client facts into a public tool, you have transmitted confidential information to a third party, and depending on the tool's terms, possibly into its training data. That is the exposure.

What breaks privilege and what does not

  • Facts about a matter, client identity, strategy, draft language tied to a client: confidential. Do not put these into a tool that transmits to a third party you have no agreement with.
  • General legal questions with no client-identifying facts: lower risk, but sloppy habits become client-facts habits fast.
  • The safest posture: assume anything you would not email to opposing counsel should not go into public AI.

How to actually use AI in practice

  1. Use private AI for anything touching a client. Private AI processes on infrastructure the provider controls and does not train on your content, which keeps the information inside a boundary you can defend.
  2. Keep a tool policy for your firm. Written, simple: which tools are approved for client matters, which are for general use only.
  3. Do not rely on consumer opt-out toggles for privileged work. A toggle is a setting; privilege deserves architecture.
  4. Verify every output. AI drafts; you are still counsel. Confidentiality is one duty; competence is another.

What to require of any AI tool before it touches a matter

  • Content is processed on infrastructure the provider controls, not silently routed to another company.
  • Your inputs are never used to train a model.
  • No third-party review of your content, and clear terms on retention.

The short version

The duty of confidentiality is absolute and technology does not soften it. Public AI transmits client information to a third party and may train on it, which is a bright-line problem for privileged work. Private AI, built so your content stays yours, lets you get the drafting and research leverage without putting the representation at risk.

Private AI that was built for this

Kiyomi runs on Jah, private AI on infrastructure Kiyomi controls. Your core chats are processed privately, never sold or shared, and never used to train a model. It is the AI you can use when confidentiality is not optional.

Try Kiyomi free — then $25/mo for everything.