AI terms

What AI is used for

Matching calls and texts to matters. When a phone number belongs to a client with more than one open matter, the model reads the transcript or messages and suggests the matter. A confident match (80% or higher, with no warning signs) is filed to the matter as the timekeeper's private draft, which they can move with one tap; anything less certain goes to the triage inbox for a person to choose.
Drafting time-entry narratives. A short description of the call or text thread, for the timekeeper to accept, edit or discard.
Reading intake calls. When the AI receptionist takes a call from a prospective client, the model pulls out the name, contact details and what they need, so staff can create a lead.

Privacy comes first

Only what the task needs. The model sees the transcript or text excerpt and the candidate matters. It never sees trust balances, invoices, documents or other clients' files.
Sensitive numbers removed first. Card numbers, Social Security numbers and bank account numbers are stripped out before text is stored or sent to the model.
No training, ever. Your firm's records are not used to train any model, by us or by our providers.
No content in our logs. We log each AI request's purpose, version, cost and outcome, not the text that was sent or returned.
Drafts stay private. Anything AI drafts is visible only to the timekeeper until they release it.
You can switch it off. With AI off for the firm, nothing is sent to the AI provider and LawMOR uses rules and templates instead.

What never happens

AI never bills. Entries drafted from calls and texts must be released by the timekeeper and approved before they can reach an invoice.
AI never sets or raises billed time. Duration is measured from the call itself, and the database refuses billed units that exceed measured time by more than one billing increment unless a person records a reason.
AI never moves trust money or signs a reconciliation.
Your clients' data is not used to train models, by us or by our AI provider.

Who provides the models

LawMOR uses Anthropic's Claude models through the Anthropic API under Anthropic's commercial terms, which do not allow Anthropic to train its models on what we send. Call audio is turned into text by Deepgram, and the AI receptionist, if a firm turns it on, runs on Vapi. All are US companies, listed with what each receives on our subprocessors page.

Safeguards around every AI call

Transcripts, messages and forms are treated as untrusted. They are clearly separated from our instructions, and a response that shows signs of following instructions found inside them is thrown away.
Every response must match a strict format; anything else is discarded rather than guessed at.
Card numbers, Social Security numbers and bank account numbers are removed before text is stored or sent for intake reading.
Every AI call is logged with the prompt version, cost and outcome.
Each firm has its own AI switch and daily and monthly spending limits. When AI is off or a limit is reached, LawMOR falls back to rules: matching by known numbers and recent threads, and plain template narratives such as "Telephone conference with Maria Delgado".

Your professional responsibility

Bar guidance on generative AI expects lawyers to review AI output and to bill only for time actually worked. LawMOR is built around that: AI drafts, you decide. This page describes the product; it is not legal advice.