Protocols for AI in legal work

JIRAF asks. JURIS verifies.

AI-assisted legal work needs protocols, not slogans. JIRAF and JURIS are a tool-agnostic pair: one frames the work before you engage an AI, the other verifies the output before you rely on it, submit it, or sign it.

Why protocols

Tools change. Professional duties remain.

Models, interfaces, and platforms change constantly, but the duties of competence, confidentiality, candour, and supervision do not. A protocol that depends on a particular tool ages with that tool; a protocol built around professional judgment survives the platform.

This is why bare instructions are not enough. "Verify the output" and "check the sources" have not worked on their own. AI-assisted legal work keeps failing in the same ways: fabricated citations, jurisdictional drift, and confident errors a quick glance does not catch. Telling someone to be careful is not a procedure. A protocol is. It establishes a defined sequence of checkpoints that holds the work to professional responsibility and recognized standards.

The logic is the same for every task: orient well before, verify well after. JIRAF and JURIS are the practitioner pair. They are not exercises in prompt engineering or context engineering. They are a layer of judgment that sits around whatever technique you use and makes the human checkpoints explicit and unavoidable.

JIRAF · how you ask

Five anchors before the first prompt.

Frame the matter before drafting with AI.

J

Jurisdiction Juridiction

Name the legal system, court, and regulatory regime explicitly, so the AI does not default to US law or carry reasoning across traditions.

I

Instruction Instruction

State a clear, specific task and the deliverable you expect. Decide what the AI may see, and use only approved tools for confidential, privileged, or client-identifying material.

R

Role & rationale Rôle

Assign the professional role and experience level, or in a designed context ask for stated assumptions, issue structure, authorities to verify, and uncertainty. AI supports judgment; it does not replace it.

A

Audience Auditoire

Name the reader and what they need. Write to that reader's expectations of structure, tone, and depth.

F

Format Format

Specify the structure, length, and register up front, and require a verification block, so you are not reshaping the answer afterward.

JURIS · how you verify

Five checkpoints before the work leaves your hands.

Verify before you rely, submit, or sign.

J

Jurisdiction Juridiction

Confirm the cited framework actually governs your facts, has not been superseded, and is the strongest available rather than merely a plausible one. Drift means re-prompt.

U

Usage Usage

Confirm the tool and the use were authorized for this matter and purpose. Do not repurpose an output into a context it was never checked against.

R

References Références

Trace every material claim. Verify that each authority is real, current, correctly characterized, and the best source for the point. Detect fabricated citations.

I

Integrity Intégrité

Read the output as a whole. Check for contradictions, reasoning that drifts, conclusions that do not follow, and material omissions.

S

Signature Signature

You remain responsible for what leaves your hands. Accountability is professional and non-delegable, and any required disclosure must be made.

Reliance levels

State what the work is ready to bear.

Pick one level before the work leaves your hands. A scale, not a disclaimer.

1

Not ready

Do not rely. Integrity issues remain.

2

Internal use only

Useful for thinking. Not for external circulation.

3

Ready after corrections

Usable once specified corrections are made and re-verified.

4

Ready for supervised use

Verified, corrected, and authorized for professional use.

The level states what the work is ready to bear. It records professional judgment; it does not replace it.

Verification note

Leave a trace of the review.

A minimal record that fits in a matter file. The note does not replace professional judgment; it records it.

Tool used
model and version
Sources checked
CanLII / SOQUIJ / statute
Jurisdiction
Ontario / Québec / federal
Claims verified
all material assertions traced
Issues corrected
hallucinations, drift, omissions
Reliance level
1 · 2 · 3 · 4 (tick one)
Reviewer / date
 

The full framework

Four roles, eight protocols.

JIRAF and JURIS are the practitioner pair. The same orient-before, verify-after logic extends to three other legal-work roles. This expanded framework is in development and open for scholarly comment.

RoleEntry · orientationExit · verification
Practitioner JIRAFFrame the matter JURISVerify before signing
Researcher / Scholar CADRESet the method frame REVUEConfirm and disclose
Professor / Lecturer TEACHDesign to the outcome ATTESTAccountability check
Student TRACEThink first, ask narrowly PROVEMake it yours to defend

AI training

Training for legal professionals, anchored in practice.

I provide AI training for legal professionals and graduate programs in Canada and Brazil, built around these protocols and around the discipline they encode: knowing what data you use and whether you may, keeping the use accountable and traceable, and verifying quality before anything is relied upon. I created the prompt and context engineering course in the ITS Rio graduate program, and I teach this material across the institutions where I hold appointments.

Beyond training, I help law firms and legal departments adapt to the evolving AI landscape and adopt the technology tools that have proven their value over the past two years, so that adoption is led by legal reasoning rather than by the tool. Sessions and engagements are tool-agnostic and adapt to your workflow and governance posture, not the other way around.

Discuss a training session
Audience
Practitioners, researchers, teachers, students
Languages
English, French, Portuguese
Format
Workshops, Continuing Legal Education, graduate seminars
Stance
Tool-agnostic. Legal reasoning-first.