Compliance & privacy

The AI never sees who your client is.
By design, not by policy.

Identifiers are stripped before the analysis model, nothing is stored, and your intake runs on your own AI account. Built to the UK GDPR standard — the strictest bar, so it holds across Australia, the UK, Canada and New Zealand — and aligned with US professional-conduct rules. The same architecture protects a law firm and a community legal centre alike.

Verified against our own code — your data-protection officer can hold us to each claim.
How it protects data

Four guarantees that hold on every intake

Not a policy you have to trust — an architecture your DPO can audit.

Guarantee 01

Identifiers stripped before the case-analysis model

Names, addresses, phone, DOB and government IDs are replaced with typed role tags — [CLIENT], [ADDRESS], [EMPLOYER]server-side, so a tampered browser can't switch it off. The reasoning model only ever works from the tagged version.

Honest note: uploaded document images are first read for text by Google Cloud Vision (an OCR sub-processor). Redaction happens before the reasoning/analysis model — the model that could retain or expose identity.
Guarantee 02

Zero client PII in our database

Our database holds only your organisation's config and a PII-free job record (session id, status, timestamps). Client names, documents and contact details are never written to it.

The only personal data we store is your own account contact — the lawyer's name and email at signup. Never the client's.
Guarantee 03

Processed transiently, deleted on delivery

Intake data lives only for the session. It is permanently deleted the moment the case file is delivered to your team. Incomplete sessions are purged within 90 days. Original documents are never retained.

Guarantee 04

BYOK — you're the controller; no training

In production your firm or centre runs on its own Anthropic (Claude) or OpenAI key. You're the data controller when you configure your key; we're a processor at most. No training on client data: on the demo path we set no-logging / no-training per request in code (data_collection: deny); production runs Anthropic-direct, and Anthropic does not train on API data.

Across five jurisdictions

Built to the UK GDPR standard — the strictest bar

Meet it and Australia, Canada and New Zealand follow. The US works through professional-conduct rules. This section is for information, not legal advice — verify for your jurisdiction; current as of July 2026.

🇬🇧United KingdomUK GDPR · DPA 2018 · SRA
  • Legal-intake data is largely special-category (UK GDPR Art. 9) and criminal-offence data (Art. 10). Because identifiers are stripped before the analysis model, that data isn't processed by the reasoning model — reducing DPIA exposure.
  • The Data (Use and Access) Act 2025 is in force — a recognised-legitimate-interests basis is available.
  • We provide a DPIA template you can adopt for your firm or centre (below).
  • Serves SRA principles on client confidentiality and public trust.
The UK sets the strictest privacy bar of the four common-law jurisdictions; building to it carries AU, CA and NZ.
🇦🇺AustraliaPrivacy Act 1988 · APPs · NSW
  • Redaction before the analysis model means personal information is not input into the reasoning system — the most direct path to OAIC's AI guidance.
  • Deletion on delivery supports APP 11 (security) and APP 4/6 (purpose limitation).
  • Covered by NSW PPIPA / HRIPA where the state scheme applies.
  • BYOK makes your organisation the APP entity / data controller — not us.
From 10 Dec 2026, automated decisions that could significantly affect a person must be disclosed in your privacy policy — a transparency duty, not a ban. Disclose any eligibility or priority scoring in your privacy notice.
🇨🇦CanadaPIPEDA · Quebec Law 25
  • Explicit informed consent at intake start meets Quebec Law 25's opt-in requirement.
  • Transient processing + deletion supports purpose limitation; no secondary use.
  • No training on client data; BYOK keeps the firm or clinic as controller under PIPEDA.
We're monitoring Ontario Bill 194 for any funded-clinic AI obligations — it is public-sector legislation and clinic scope is uncertain.
🇳🇿New ZealandPrivacy Act 2020 · IPP 3A
  • The widget tells clients at the start exactly what happens to their data — supporting the new indirect-collection notification, IPP 3A (in force 1 May 2026).
  • Collection limitation, purpose specification and deletion-after-delivery align with the 13 IPPs of the Privacy Act 2020.
  • NZ integrates AI expectations into existing law — no separate AI Act.
The least prescriptive of the four — satisfied once AU and UK are.
🇺🇸United StatesABA Op. 512 · Model Rules · State bars
  • ABA Formal Opinion 512 (July 29, 2024): before using AI with client information, lawyers must understand how the tool handles data, whether inputs train the model, and the disclosure risk — and obtain informed client consent. PROVEAiBLE is built around those requirements.
  • The ABA Task Force Year 2 Report on the Impact of AI on the Practice of Law (Dec 2025) echoes and continues those obligations.
  • Model Rule 1.6 (confidentiality) + Model Rule 1.18: the duty to a prospective client attaches at intake, before representation forms — a site inviting confidential information creates the duty from the first message. Intake is protected from the moment of capture.
  • Consistent state-bar guidance: the NYC Bar (Formal Ops 2024-5 & 2025-6 — the latter concerns AI recording/transcribing client conversations), Texas Opinion 705 (Feb 2025), Florida Opinion 24-1 (Jan 2024), and the Illinois Supreme Court AI Policy (Jan 1, 2025 — a court-use policy).
  • California CCPA / ADMT: compliance required from Jan 1, 2027; legal intake and triage sit outside the regulation's enumerated "significant decision" scope.
  • HIPAA: most plaintiff-side PI and immigration firms are not HIPAA business associates and their intake data isn't PHI; defense-side or insurer-retained firms may need a BAA.
No single federal privacy law governs legal intake; the framework is professional conduct and evolving state regulation. We update this page when material changes occur.

For information only — not legal advice. Verify the position for your jurisdiction. Current as of July 2026.

For DPOs

Sub-processors

Five providers touch an intake. Here's each one — what it sees, and whether it keeps anything. Only the OCR step ever reads a raw upload, the model that analyses the case sees only the redacted version, and the one place full client details appear is the finished notification we email to your firm — because you're the one receiving the matter. All bound by data-processing agreements.

RenderApplication hosting
SeesThe intake as it runs — held only in memory while processing.
KeepsNothing — wiped on delivery.
Google Cloud VisionDocument OCR
SeesRaw text from uploaded files — read once so it can be redacted.
KeepsNothing.
Anthropic / ClaudeAI analysisRedacted only
SeesThe redacted version of the intake — never raw identifiers. Runs on your BYOK key or a managed Anthropic-direct key.
KeepsNothing — Anthropic does not train on API data.
SupabaseDatabase
StoresYour firm's setup, and the status of each intake — which matter, what stage, who it's assigned to.
NeverThe client's name, contact details, or case content.
ResendEmail delivery to your firm
SeesThe finished notification sent to you — including the client's contact details and case summary.
KeepsNothing — delivery only.
Take it to your board

DPIA template & compliance checklist

The detailed, dated legal mapping — current as of July 2026. Enter your email and we'll send it straight to your inbox. Verify for your jurisdiction; not legal advice.

UK DPIA template (intake AI)

A Data Protection Impact Assessment pre-filled for AI client intake — adapt it to your firm or centre and file it.

Compliance checklist (AU · UK · CA · NZ)

A one-page checklist per jurisdiction — map PROVEAiBLE to your obligations and tick them off. US firms: a reference mapped to ABA Formal Opinion 512, Model Rules 1.6 and 1.18, and 2025–2026 state-bar guidance, available on request.

We'll email you the document; we may follow up once. No list-selling, ever. This page is information about our architecture, not legal advice — your firm or centre remains bound by its own privacy and professional-conduct obligations.

Questions

Common compliance questions

Does the AI ever see our clients' personal details?

No. Names, addresses, document IDs and the like are stripped to typed role tags on our server before the case-analysis model — the reasoning model that could retain or expose an identity. It only ever works on the redacted version, and this is enforced in the architecture, not an optional setting. Uploaded document images are first read for text by an OCR sub-processor (Google Cloud Vision); redaction happens before the analysis model.

Do you train AI models on our clients' data?

No. On the demo path we set no-logging / no-training per request in code (data_collection: deny), not just as an account setting. In production the intake runs on your organisation's own AI account (BYOK) via Anthropic direct, and Anthropic does not train on API data. No client PII is written to our logs.

Where is our clients' data stored, and for how long?

We don't retain client matter data. It's processed transiently to build the case file, then permanently deleted the moment that file is delivered to your team; incomplete sessions are purged within 90 days. The only things we store are your organisation's own configuration, a PII-free job record (session id, status, timestamps), and your own account contact (the lawyer's name and email at signup) — never the client's. Operational infrastructure runs through named sub-processors bound by data-processing agreements.

Who is the data controller — you or us?

In production the intake runs on your organisation's own AI key (BYOK), so you are the data controller when you configure your key. For the brief, transient intake step we act only as a processor; once the case file is delivered it lives with your team. We never become the controller of your clients' matter data.

Can we show that the AI never saw identifying information?

Yes — by design, not just by assurance. Every identifier the client confirms redacted is captured with a timestamp as a consent record (GDPR Article 7 demonstrability) and delivered alongside the case file. The token-to-name mapping is never sent to the AI and is deleted on delivery — so the delivered file is itself evidence the analysis model only ever processed the protected version. Using a consumer AI chatbot can waive privilege, which is exactly why PROVEAiBLE strips identifiers before the analysis model.

See the architecture run on your matter type

A private demo tuned to your practice — with the redaction, consent record and case-file output your DPO can review.

Request a demo