Artificial Intelligence Use and Human Review Policy
How artificial intelligence may and may not be used across the firm and its systems. Version 1.0.
This policy governs how artificial intelligence is used in NRI Legal, its associate practice, its CRM and CMS, and any tool that touches client information. It reflects a promise we make publicly and contractually: client information never goes to public artificial intelligence, and a qualified person approves anything before it is sent or filed. It is also an information security control for ISO 27001 and a requirement of our data protection obligations.
The two classes of artificial intelligence use
Class one, generation. Drafting documents and communications. This is deterministic: the system fills approved templates with confirmed data from the client record. It does not invent legal content. Every generated document or message is reviewed and approved by a qualified person before it is sent to a client or filed anywhere. Generation is never fully automatic to the outside world.
Class two, read and summarise. Internal assistance such as summarising a matter, extracting facts, or surfacing information for a person to act on. Output is for internal use, is treated as a draft, and is checked by a person before it is relied upon. It never becomes an outbound communication or a filing without human approval.
The absolute rules
Client and prospect personal data is never entered into, processed by, or disclosed to any public or third party artificial intelligence service.
Client personal data is never used to train any artificial intelligence model.
Any artificial intelligence used with client data operates only within our own controlled systems.
A qualified person always reviews and approves artificial intelligence output before anything is sent to a client, filed with a court or authority, or acted upon.
Artificial intelligence never exercises legal judgment, never gives advice, never assesses the other side, and never approves a document for filing. Those are always human.
The quality gate on outbound work
All outbound communications and documents pass an artificial intelligence assisted quality check before human approval. The check does two things: it checks quality and tone against the firm benchmark, and it checks data accuracy by comparing against the master client record so that manual entry errors are caught. The check is a review layer only. It does not send anything. Generation stays deterministic from templates, and a person always approves before anything leaves the firm.
Data accuracy by design
Data is never typed manually into documents where it can be lifted from the master client record. Pulling data from the single record, rather than re-keying it, is what makes entry errors structurally unlikely. Anyone building or using the system follows this rule.
Records and oversight
The system logs artificial intelligence assisted actions and the human approval that followed, so we can show who approved what and when. [Named owner] is responsible for this policy, for approving any new artificial intelligence tool before it touches client data, and for reviewing this policy at least annually and whenever a new tool or regulation requires it.
Applies to the build
This policy binds the development of the CRM and CMS as much as its daily use. Any feature that generates or sends must keep human approval in the loop, and no build or test activity may place real client data into a public artificial intelligence service.
[Confirm the named policy owner.]