Confidentiality needs architecture and judgement

Private legal AI that can show where an answer came from.

Legal teams handle privileged, personal and commercially sensitive material. A private route can reduce external disclosure, but professional judgement, matter permissions and governance remain essential.

Decision in one minute

Start with an internal, low-risk research or knowledge workflow. Do not automate legal advice or client decisions without appropriate controls.

A private deployment starts with the permitted data path, access policy and logging boundary.

Sector fit

Start with duties and data boundaries, then test one useful workflow.

01 / Duty: Map confidential data, professional duties and approvers first.
02 / Pilot: Use one bounded workflow and representative, permitted material.
03 / Control: Retain human review, access ownership and an evidence trail.

Diagram showing approved documents moving through a searchable index to an answer with a source citation — A retrieval workflow should connect each useful answer to approved source material and defined refusal behaviour. Original explanatory plate. It sets out a decision method, not a measured result.

Good first workflows

Internal precedent and policy search, chronology support, document classification, first-pass summaries and drafting assistance can be tested against source evidence.

The output should cite the matter-approved material and remain subject to qualified review.

Assist Search, extract, classify or draft.
Cite Show the evidence used where applicable.
Review Authorised person checks the result.
Decide Accountable role accepts or rejects.

The exact checkpoint depends on the workflow, risk and organisation’s professional obligations. Decision aid for: Good first workflows

GPU server remote management dashboard with system status, access logs and sensor monitoring panels — Supplier screenshot of the platform management interface. The final management features and access policy depend on the ordered system. OEM supplier reference image. Written reuse permission pending.

Diagram of an evidence pack containing an asset schedule, burn-in record, health readings, workload test and admin guide — A credible handover records the supplied assets, checks, operating evidence, instructions and unresolved items. Original explanatory plate. It sets out a decision method, not a measured result.

Matter-level permissions

An index must not flatten ethical walls or matter restrictions. Retrieval needs group-aware access or separate stores.

User identity and access reviews remain customer responsibilities.

Reference pattern: Reasoned but not customer proof.
Bounded pilot: Permitted material and pass conditions.
Witnessed result: Conditions and reviewer recorded.
Permission: Publication scope agreed in writing.

Concept diagrams and supplier images never become evidence of a customer outcome. Decision aid for: Matter-level permissions

Accuracy and privilege

Fluent output can be wrong. Evaluation needs refusal, citation and escalation criteria.

Local hosting does not by itself preserve privilege or satisfy SRA, client or cyber-insurer expectations.

Owner Data, process and technical roles named.
Material Representative and permitted scope.
Boundary Decisions the service may not make.
Expansion Evidence required before wider use.

A successful pilot is permission to evaluate the next boundary, not proof that every sector workflow fits. Decision aid for: Accuracy and privilege

A bounded pilot

Use a representative, de-identified or authorised document set and a written question set. Record answer support and reviewer effort.

Scale only when the quality and governance case survives.

Questions answered

Straight answers to common questions

Does private AI protect legal privilege?

Architecture can reduce external disclosure, but privilege depends on legal and operational circumstances. Obtain legal and professional guidance.

Can it draft legal documents?

It can assist with approved drafting workflows, but every material output needs qualified review and source checks.

Is the system SRA approved?

No such claim is made. A firm must assess its professional, data-protection, security and client obligations.

Primary-source register

Check the live rule or price before relying on it.

Reviewed 26 July 2026. These links support the dated statements on this page; they do not replace legal, tax, security or professional advice.

SRA 2026 AI risk outlook Current sector context from the regulator for England and Wales. ICO: AI and data protection Data-protection guidance does not replace legal-sector professional duties.

Continue the decision

Useful next steps

Put the claim to work

Turn this guidance into a testable requirement.

The brief asks about workload and operating conditions - not just budget.

Plan your requirements Compare proposed systems