World Model Readiness
Engraved team-functions instrument

For your team

Teams by Function

Module · AI that does not waive privilege

The Legal Team AI Check

Legal work has a property most work does not: the confidentiality of the material is itself a legal protection, and it can be lost by a single careless act. AI is enormously useful to a legal team and enormously good at destroying exactly that protection. This module checks the five things that let your team use AI without giving away its advantages: keeping privilege intact, verifying what the AI says about a contract, catching the citations it invents, holding the confidentiality boundary between matters, and capturing what the team learns so the AI does not have to relearn it every time.

Question 1 of 5 · Privilege stays intact

Do you know whether feeding privileged material to an AI tool risks waiving privilege?

Legal privilege depends on confidentiality, and pasting privileged material into a tool that trains on inputs or shares them with a provider can be argued to breach it. The protection that makes legal advice safe to give can be waived by one lawyer reaching for a convenient chatbot. This is a question you answer before use, not after a dispute.

Question 2 of 5 · Contract reads are verified

When AI reviews or summarises a contract, does a lawyer verify what it claims the document says?

AI is confident and often wrong about what a clause actually does: it misses a carve-out, inverts an obligation, or reassures you about a risk that is right there in the text. A summary taken on trust is worse than no summary, because it stops the lawyer from reading the clause themselves. The document is the authority, not the AI's account of it.

Question 3 of 5 · Citations are checked

Does someone check that the cases and clauses the AI cites actually exist and say what it claims?

AI invents citations that look perfect: real-sounding case names, plausible section numbers, confident quotations, none of it real. Lawyers have been sanctioned for filing them. Every authority an AI provides has to be pulled and read at the source before it goes anywhere near a court, a client, or a contract.

Question 4 of 5 · Matters stay walled

Do AI tools respect the confidentiality walls between clients and matters?

Legal teams keep information barriers between clients, sometimes between teams on the same matter. An AI tool that pools every document it has seen, or lets one matter's context surface in another, drives straight through those walls. A conflict of interest created by a shared AI index is no less a conflict for being accidental.

Question 5 of 5 · Precedent is captured

Does the team capture its own precedents and positions so AI works from them, not from scratch?

A legal team's value is accumulated judgement: the clauses it prefers, the positions it has taken, the playbook it has built. If AI drafts from the open internet every time instead of from your own captured precedent, you get generic work and you relearn the same lessons repeatedly. Captured knowledge is what turns a general model into your team's tool.

For the statistics · one click each

Three questions for the public picture

These do not affect your score. They feed the anonymised, aggregated statistics; groups under 8 respondents are never shown.

Where does your legal team use AI today?

Not yet
Legal research
Contract review
Drafting and negotiation
Across most matters

Do you have a rule for privileged material going into AI tools?

No rule
Assumed safe, no rule
Informal caution
Written policy, vetted tools
No AI used on legal work

How does the team handle citations an AI provides?

Used unchecked
Names spot-checked
Checked before filing
Every citation pulled and read
AI not used for authorities

Your context

Used to calibrate the report. Company size and sector remain in the anonymized dataset; your email does not.

What the five levels look like

Every dimension in this assessment is scored 1 to 5. This is what the levels mean, dimension by dimension. The graded report diagnoses where your own answers land and what to do about it.

Privilege stays intact

  1. 1Never considered
  2. 2Assumed fine
  3. 3Aware, no policy
  4. 4Policy on privileged data
  5. 5Vetted tools only

At the low end: If nobody has asked whether your AI tools waive privilege, assume the risk is live and act on it. Pause privileged material going into any tool until you know how it handles and retains inputs. What good looks like: Restricting privileged work to vetted tools with contractual confidentiality is how you keep the protection intact. Re-vet as terms of service change; a provider that did not train on inputs last year may quietly start.

Contract reads are verified

  1. 1Summary trusted
  2. 2Skimmed against text
  3. 3Key clauses checked
  4. 4Verified against document
  5. 5Verified and annotated

At the low end: Trusting an AI summary of a contract without checking the text is how a missed carve-out becomes signed. Require a lawyer to verify every AI claim against the actual clause before relying on it. What good looks like: Verifying AI claims against the document itself keeps the contract, not the model, as the authority. Keep the habit strict on the clauses that carry the risk; that is exactly where a confident summary is most dangerous.

Citations are checked

  1. 1Cited unchecked
  2. 2Names spot-checked
  3. 3Checked when filing
  4. 4Every citation pulled
  5. 5Pulled and read

At the low end: Using AI-supplied citations without verifying them risks filing authorities that do not exist. Pull and read every case and clause the AI cites at the source before it leaves the team. What good looks like: Pulling and reading every AI citation at the source is the only safe standard, and it is the one courts now expect. Keep it absolute; the one unchecked citation is the one that ends up in the sanctions order.

Matters stay walled

  1. 1One shared pool
  2. 2No barriers
  3. 3Manual separation
  4. 4Access-scoped by matter
  5. 5Scoped and audited

At the low end: An AI that pools every matter into one index can surface one client's confidential material inside another's work. Scope the tool's access by matter before that becomes a conflict you have to disclose. What good looks like: Access scoped by matter and audited keeps the information barriers intact when AI enters the workflow. Review the scoping as teams and tools change; a barrier is only as good as the last configuration nobody checked.

Precedent is captured

  1. 1Nothing captured
  2. 2In individual heads
  3. 3Scattered folders
  4. 4Structured precedent base
  5. 5Curated and AI-connected

At the low end: Without captured precedent, your AI drafts from generic sources and your team relearns its own positions each time. Start collecting your standard clauses and past positions in one place the AI can draw on. What good looks like: A curated precedent base connected to your AI tools turns a general model into an instrument that reflects your team's judgement. Keep it curated; an unmaintained precedent base teaches the AI last year's positions.