Trust Centre · v1.0
Sovereign by default. Evidence-ready by design.
This is the plain version — for the procurement officer who has read a hundred glossy compliance pages and just wants the facts. Where your data lives. Who can touch it. What we can prove today. How to get the full pack.
Last updated 17 June 2026
Your data stays close
Your data lives in Sydney today, not the US. An NZ-resident option is on the way. Only masked content ever reaches a model.
Models never see real names
We strip the personal details before any model call. The model reads masked text. The names stay with you.
Proof you can hand over
Every output ends in an evidence pack — a downloadable bundle of PDFs. Hand it to an auditor and they can check it themselves.
Architecture
The Mana Trust Layer
Last updated 17 June 2026
Every output runs the same five steps: Kahu → Iho → Tā → Mahara → Mana. Kahu takes the request and masks the personal details first. Iho picks the right agent and model. Tā drafts the work with every source cited inline. Mahara is where someone in your team reads it and decides. Mana seals the receipt with their name on it. The personal details are masked before any model call — the model never sees a real name.
Where your data lives
Data residency
Last updated 17 June 2026
Your data lives in Sydney today. Not in the US. It is the closest major cloud region to Aotearoa, and an NZ-resident option is on the way — we are confirming the region with our infra team now. Models may sit offshore, but they only ever see masked content. Your raw data stays put.
assembl
Your data sits in Sydney, not the US.
An NZ option is on the way. Everything is fenced to your tenant, and only masked content ever reaches a model.
The implicit alternative
Routed offshore by default.
Most AI tools ship your data to the US the moment you paste it. No regional option. No masking. The model sees everything.
Who touches your data
Sub-processors
Last updated 17 June 2026
Here is every company that can touch your data — where they sit, what they do, and the contract that binds them. We check this list every week. Every change shows up in the log below.
| Provider | Location | Purpose | Data | DPA |
|---|---|---|---|---|
| Anthropic (Claude)Masked content only | United States | Primary model inference for draft generation. | Workflow content | View DPA |
| Google (Gemini)Masked content only | United States | Model inference for selected workflows. | Workflow content | View DPA |
| OpenAIMasked content only | United States | Model inference for selected workflows. | Workflow content | View DPA |
| Supabase | Australia (Sydney) — NZ-resident option in progress | Primary database, authentication, and file storage. Row-level security per tenant. | Workflow content, Account metadata, Operational logs | View DPA |
| Vercel | United States / global edge | Application hosting and content delivery. | Operational logs | View DPA |
| Stripe | United States | Billing and payment processing. | Account metadata | View DPA |
| Twilio | United States | Transactional SMS and notifications. | Account metadata | View DPA |
What we can prove today
Compliance posture
Last updated 17 June 2026
What is true today, not what we hope for. If a certification is not done, we say so. We would rather tell you we are on the way than pretend we have arrived.
Privacy Act 2020 (NZ)
LiveThis is the law your data actually lives under. We run to the Privacy Act 2020, including the new IPP 3A rules from 1 May 2026. Our Privacy Statement spells out what we collect and how to complain.
NZISM alignment
In progressWe line our controls up against the NZ Information Security Manual. It is our own honest read, not a government tick.
SOC 2 Type 1
PlannedNot done yet, and we won't pretend otherwise. We are scoping a SOC 2 Type 1 readiness check. The moment we engage an auditor, their name and the date land right here.
Auditor and date: coming.
ISO 27001
PlannedNot certified. It is next after SOC 2. We won't claim it until an auditor has signed off — full stop.
Lined up after SOC 2.
Verifiable by anyone
Evidence packs
Last updated 17 June 2026
Every output ends in an evidence pack — a downloadable bundle of PDFs. Sources, decisions, who signed off, and when. Show it to a regulator, an auditor, or a client, and they don’t have to take your word for anything.
Under the hood we call it a Mana Receipt. It’s Ed25519-signed, so any tampering shows. Paste one into our public verifier and check it yourself.
How it's protected
Encryption
Last updated 17 June 2026
At rest
Stored data is encrypted with AES-256. A stolen disk reads as noise.
In transit
Every connection runs on TLS 1.3. Nothing travels in the clear.
Key management
Our infra providers hold the keys, kept apart from the data they unlock. The app code never sees them.
When something goes wrong
Incident response
Last updated 17 June 2026
Detect
Logs and audit trails flag access that looks wrong, fast.
Contain
We kill the affected access first, then map exactly how far it reached.
Notify
We tell you — and the Privacy Commissioner when the law says we must.
72 hours, not “eventually.” The Privacy Act 2020 asks for notice “as soon as practicable.” We hold ourselves tighter: you and the Privacy Commissioner hear from us within 72 hours of us knowing.
Your data, your terms
Deletion & retention
Last updated 17 June 2026
You decide how long we keep things. By default, evidence packs stay as long as your contract needs — usually to match your own tax or audit rules. Ask us to delete your data and we do, unless a law makes us hold it.
- • By default: kept as long as your contract and record-keeping rules need.
- • Your call: dial the window up or down per workspace.
- • Deletion: ask, we remove it, then confirm it’s gone.
Who can see what
Access controls
Last updated 17 June 2026
Row-level security
Every record is fenced to its tenant. One customer simply can't read another's.
Role-based access
Inside your workspace, people see what their role allows — and nothing past it.
Founder allowlist
The riskiest outputs wait behind a named person. Nothing sensitive ships without them.
How the agents are governed
AI governance
Last updated 17 June 2026
The agents draft. People decide. Te Tiriti commitments and tikanga values shape how we build them and how they handle your data.
- Draft-only. No agent sends, posts, or files anything on its own.
- A named person reads every draft and accepts, edits, or bins it.
- Nothing leaves your workspace until someone signs it off.
- Your data never trains a public model. Ever.
Last 12 months
Change log
Last updated 17 June 2026
- 17 Jun 2026First publication of this Trust Centre and the sub-processor list.
Get the full pack
Request the security pack
Last updated 17 June 2026
Running a vendor review? Tell us who you are and what you need. Our security team replies straight from security@assembl.co.nz.
Kate Hudson, our founder, reads and signs off this page herself. See something that’s not right? Email security@assembl.co.nz.