The AI Stop/Go Framework
A Decision Map for Lawyers. Four risk factors, a Stop/Go reference table, a worked example on settlement advice, and a quick-reference card for deciding when AI is appropriate in legal work.
A Decision Map for Lawyers
The Core Principle
AI is a capable drafting, structuring, and summarising tool. It is not a professional decision-maker. The boundary between the two is not always obvious — and the consequences of blurring it fall on you, not the model.
This framework gives you a structured way to make that judgment call before you begin.
The Four Risk Factors
Before using AI on any task, run it through these four questions:
| Risk Factor | Question to Ask | If YES |
|---|---|---|
| Fact-intensive | Does the answer depend heavily on specific client facts? | Human judgment required for conclusions |
| Jurisdiction-sensitive | Does this area of law change frequently or vary by jurisdiction? | Verify currency before relying on output |
| Emotionally complex | Could poor handling cause distress or erode client trust? | Human drafting or heavy review required |
| Confidentiality risk | Does the task require inputting client-identifying information? | Use only approved enterprise tools with DPA |
The Stop/Go Reference Table
| Task Type | Signal | Recommendation |
|---|---|---|
| Drafting standard letters / emails | Low fact-intensity, common format | GO → use AI, review output |
| Explaining legal concepts in plain English | No client-specific facts needed | GO → ideal AI task |
| Summarising documents you have read | You control what goes in | GO → efficient use of AI |
| Generating issues checklists | Structured, low-risk output | GO → use as starting point |
| Citing cases or statutes | High hallucination risk | VERIFY → never use unverified |
| Advising on contested facts | Requires professional judgment | HUMAN → AI can draft; you decide |
| High-emotion matters (family, criminal) | Client relationship at stake | HUMAN → review carefully |
| Advice on fast-changing law | Model may be out of date | VERIFY → check live sources |
| Inputting client PII into free tools | Data protection risk | STOP → use approved tools only |
| Final advice to client | Professional accountability | HUMAN → AI output is a draft only |
Worked Example: Settlement Advice
Scenario: A client asks whether they should accept a settlement offer.
High-risk use: Pasting the facts into AI and asking “Should my client accept this?”
Why: The AI has no access to the client’s risk appetite, the opposing counsel’s track record, the judge’s tendencies, the full evidentiary record, or any of the hundred contextual factors that make this a professional judgment.
Better use: Ask the AI to “Structure the key considerations a lawyer should weigh when advising on whether to accept a settlement in a commercial dispute.” Use the framework it generates as a checklist for your own analysis.
The Liability Reminder
AI tools do not hold practising certificates. They cannot be sued, struck off, or disciplined.
If an AI-assisted work product causes harm to a client, the liability analysis will focus on the professional who supervised and signed off the work — not the tool that produced the first draft.
The question is never “Did AI get this wrong?” The question is always “Did you catch it?”
Quick-Reference Card
Before using AI on any task, ask:
- Is this fact-sensitive? YES = you decide; AI drafts only.
- Is the law current? YES = verify on a live source.
- Is a client relationship at stake? YES = human tone and review.
- Is client data involved? YES = use approved tool only.
- Will this go to client or court? YES = full professional review required.
Part of the AI Foundations for Lawyers series.