Skip to content
Trust · privacy · compliance · security

Trust. Privacy. Compliance.

ezylegal.ai is APP-compliant by design, AU-resident on request, and the only AU AI legal platform with cross-matter quarantine and a signed AI audit trail. This page is the dry, factual one.

Privacy Act 1988 (Cth) + APPs.

ezylegal.ai is APP-compliant by design. Personal information is handled per APPs 1 through 13: open and transparent management, anonymity options where reasonable, collection minimisation, use-limitation, integrity, and access/correction. We maintain a Privacy Officer and respond to access/correction requests within statutory windows.

Notifiable Data Breaches (NDB) scheme: any eligible data breach is assessed against the serious-harm threshold; where reportable, OAIC and affected individuals are notified within 30 days.

Australian data residency.

Default storage is AU-region across our infrastructure providers. Sydney-region origin is available on request for firms with formal residency requirements. Files at rest, the Postgres primary, and the embedding vector store all live in AU.

M365 mailbox content is read via Microsoft Graph in real time and never leaves the customer's Microsoft tenant. We persist parsed metadata (subject, sender, AI analysis blob) — never the raw mailbox.

Cross-matter quarantine.

Files and facts are isolated per matter. Rachel and ezyBarrister can only ever retrieve evidence belonging to the matter currently in scope — by construction, not by policy. If something goes wrong (a privileged file lands in the wrong matter, a client sees an internal note, a draft email is queued to the wrong recipient) it's logged as a Privacy Incident and quarantined immediately.

Privacy incident kinds

Kind Severity Description
Cross-matter contamination High → Severe A file or fact from matter A surfacing in matter B. Detected automatically by the per-matter isolation enforcement; quarantined and reviewed.
Accidental client visibility Medium → High An internal note or unredacted draft visible to a client. Logged, escalated, remediated.
Wrong recipient send Medium → Critical Mail sent to the wrong party. Tier-3 outbound actions require explicit approval to reduce the surface area of this class of incident.
Other Low → Critical Catch-all category for anything else flagged. Reviewed by a human; escalated when relevant.

Status flow: Open → Contained → Under Review → Resolved, with a separate Reported to OAIC branch for incidents that meet the NDB-reporting threshold. Affected files are auto-quarantined and excluded from retrieval; remediation is logged.

Signed AI audit trail.

Every action Rachel takes is logged. Every prompt ezyBarrister runs is logged. Every AI generation is reviewable per matter, per lawyer, per firm. The audit log is append-only and content-addressed — entries can't be silently rewritten.

  • Action

    The Rachel tool that ran (e.g. extract_facts, draft_email, classify_inbox).

  • Actor

    Rachel + the lawyer who has the matter, or the lawyer who ran the action manually.

  • Inputs

    The matter, the file or email, the prompt, the case-context fingerprint.

  • Output

    The full text of what Rachel produced.

  • Approval state

    Auto · Pending · Approved · Rejected · Reverted.

  • Timestamp

    AEDT, with original UTC retained.

Tier-3 actions (anything outbound to opposing counsel, anything destructive of evidence, large-volume client comms) carry an additional Approval state field with the reviewing lawyer's identity. The chain "Rachel proposed → Lawyer approved → Action executed" is auditable end-to-end.

Encryption at rest, in transit.

TLS 1.3 in transit. AES-256 at rest. Per-customer encryption keys for file storage on request. The vector store storing fact embeddings sits behind the same key envelope as the source files. Backups are encrypted with separate keys; restore procedures are tested quarterly.

SSO. MFA. SCIM on request.

For firms: SAML 2.0 and OIDC against your IdP (Microsoft Entra, Okta, Google Workspace). MFA is enforced by default; trust accounting and AI generation are 2FA-gated. SCIM 2.0 provisioning is supported on request for enterprise plans.

For solos: Microsoft / Google sign-in or email-magic-link with TOTP.

Compliance posture.

  • Privacy Act 1988 (Cth) + APPs Compliant by design
  • Notifiable Data Breaches scheme OAIC-reporting workflow
  • AU professional conduct rules (LSB / LCA) Aware; firm-side responsibility
  • ISO 27001 Roadmap (target: 2026)
  • SOC 2 Type II Roadmap (target: 2026)
  • Professional indemnity Customer-side; we do not advise

For a current security pack (SIG-Lite, CAIQ, pen-test summary under NDA, sub-processor list) — email hello@ezylegal.ai or use the contact form.

FAQ

Frequently asked questions

Does Rachel see other firms' data?
No. Per-matter isolation is enforced at the data-access layer — Rachel only ever has visibility into the matter she's currently working on. Cross-firm data is hard-walled. Cross-matter contamination within a single firm is a privacy-incident class with quarantine workflow, not a normal occurrence.
Where is our data physically stored?
AU region by default. Sydney-region origin available on request for firms with formal residency requirements. Mailbox content remains in the customer Microsoft tenant — Rachel reads it via Graph API in real time and stores parsed CaseEmail metadata + AI analysis, not the raw mailbox.
Can we export everything if we leave?
Yes. Full case export at any time — pleadings, brief paragraphs, facts, evidence, files, steps, messages, audit log. Data is yours; the platform is rented.
How do we satisfy our internal security questionnaire?
Talk to us. We maintain a current SIG-Lite and a CAIQ; we'll work through your firm's specific questionnaire. SSO/SAML/OIDC is available; SCIM provisioning on request. Penetration test reports available under NDA.
What about AI hallucinations in legal drafting?
Every Rachel and ezyBarrister output is grounded in the matter file (RAG over your facts, brief, pleadings, files). Citations point to the source page in the source document. The lawyer is always the final reviewer — Tier-3 outbound actions require explicit approval. The signed audit trail makes "who said what when" reconstructible after the fact.

Take the first step

Run our security questionnaire against ours.

Send your firm's vendor questionnaire and we'll respond with the relevant artefacts. Most are answered within five business days.