01 · Where the data goes

Three places, and one line between them.

In your office

Your staff's browsers

Fee-earners and support staff use a browser. Nothing is installed on their machines.

In your office

One Windows machine

Runs AI-Switchboard under a dedicated service account. Roles, guardrails, secret scanning, verification gates and the review gate all run here. Configuration, the audit database, transcripts and skills sit in one folder, which you back up to your own target.

Outside your office

The Anthropic API

Each run starts one headless model process, which sends that run's prompt over HTTPS and receives the reply. This is the only traffic that leaves.

AI-Switchboard makes no outbound calls of its own. No telemetry, no phone-home, no remote access.

02 · The model provider

Your firm contracts with Anthropic directly.

Your firm holds its own Anthropic account under Anthropic's commercial terms. For EEA and UK firms the contracting entity is Anthropic Ireland, Limited. Anthropic does not train on API inputs under those terms, and retains request content for 30 days, with a trust-and-safety exception.

Because the firm holds the account, AI-Switchboard is not in the data chain for model traffic and there is no sub-processor list for it to publish.

03 · Support

We see your data only when you ask us to.

If you ask for help reading your audit database or transcripts, AI-Switchboard sees that content. This is its only role as a processor, and it is covered by a support-only DPA within your terms. Nothing is processed without a request from you.

04 · Enforced controls

What the software does, whether or not anyone remembers to.

Review gate, on by default

The Word export refuses unreviewed runs. A named fee-earner records the review, and that name appears in the footer of the released document.

Verification gates

A skill can declare a machine check. If the check fails, the run fails, whatever the model reports about its own work.

TO CONFIRM markers

Drafting skills are instructed to flag unsupported facts, and the Word export highlights them for the reviewer. An instruction is not a guarantee.

Secret scanning

Credentials that appear in output are redacted in storage and block the export.

Audit trail

Every completed run is logged, whatever started it: dashboard, API, scheduler or command line. The record includes skill, model, user, verification result and reviewer.

Named users and roles

In firm-server mode each user has a named account. Admin actions and ad-hoc runs are admin-only.

Drafting only

The product does not check authorities against a legal database. Citations and legal assertions must be verified by the named reviewer.

05 · Law Society of Ireland guidelines, December 2025

How the controls line up with the guidance, and where each one stops.

This is not Law Society approval, and installing the software does not make a firm compliant. It is a map to help your firm do its own assessment.

Area of the guidelinesWhat the product doesWhere it stops
Duty of Confidentiality and PrivilegeOn-premises installationIt does not mean all data stays on site: each drafting request's text is sent to Anthropic and the reply comes back.
Confidentiality and Security; Duty of Confidentiality and PrivilegeFirm-owned Anthropic accountThe firm must assess Anthropic's terms and the information sent in each request; an account alone is not a safeguard.
Accuracy and understanding; Independence; Duty to Supervise; Duty to not mislead the courtNamed reviewer gateRecording a review does not prove that every statement was checked.
Hallucinations; Duty to not mislead the courtTO CONFIRM markersThe model can miss a marker: all facts, authorities and legal assertions still need independent verification.
Duty to Supervise; ConclusionAudit record of completed runsA run aborted part-way does not yet write an audit row, and a log is not proof that a draft is accurate.
Gen AI in legal practice: Opportunities and RisksDraft-only outputA draft may still contain an error; restricting its intended task does not verify its content.
Gen AI in legal practice: Opportunities and RisksNo autonomous legal adviceThe firm's people must exercise that judgement themselves.

Source: Law Society of Ireland, Guidelines for the Use of Generative AI by the Legal Profession in Ireland (v4, December 2025)

06 · The governance pack

The paperwork comes with the seat.

Documents your firm can adapt and keep on file. They support your own assessment and do not replace it.

  • Plain-language data-flow summary
  • Pre-filled DPIA
  • Register-of-processing entry
  • EU AI Act deployer checklist, with an AI-literacy record
  • AI use policy template, in WordDrawing on the Law Society of Ireland, SRA, Law Society of Northern Ireland and judiciary guidance.

07 · Reporting a concern

Email security@ai-switchboard.net

Mark the subject URGENT if client confidentiality may be breached. Good-faith reports are welcome and no one is pursued for making one. There is no paid bounty.

  • AcknowledgementWithin 2 business days
  • Initial assessmentWithin 5 business days
  • Updates on confirmed issuesAt least weekly, or at once on a material change

08 · What we cannot do

The other side of keeping us out.

See your data

There is no telemetry and no remote access.

Detect incidents on your server

Monitoring is a control on your side.

Restore your data

Backups are yours, to your own target.