# Family Office AI That Doesn’t Train on Your Data — Super Chain

> Super Chain does not train on your data because it runs inside your own infrastructure. Here is how that architecture keeps family office data in the room.

Source: https://www.super-chain.tech/trust  
Last modified: 2026-10-23

trust, answered by architecture

# *"Does it train* on our data?"

Every cloud vendor answers this question with a policy: *we promise not to.* We answer it with architecture: the platform runs inside your walls, so there is no vendor-side model to train, no shared tenant to leak into, and no retention policy to read. *Nothing leaves the room.*
no vendor-side model · no other tenants · no policy required Written by [Cat Yung](https://www.super-cat.tech/cat-yung), Founder, Super Chain. Last updated 23 October 2026.

policy vs. architecture

## A promise you read, or *a wall you can inspect.*

The concern is well founded — RSM, J.P. Morgan and Citi have each warned family offices about what AI tools may quietly learn from the data placed in them, and in July 2026 Squire Patton Boggs named confidentiality the sector’s defining AI constraint. [source: SPB White Paper, Jul 17 2026 →](https://www.familywealthreport.com/article.php/--Managing-Family-Offices%27-AI-Risk-%E2%80%93-SPB-White-Paper?id=208280) M22 and Plante Moran now publish the same warning as practical guidance for principals asking how to use AI safely with confidential data. [source: M22 →](https://www.m22.group/answers/family-office-ai-confidential-data) [source: Plante Moran, Mar 2026 →](https://www.plantemoran.com/explore-our-thinking/insight/2026/03/innovating-with-ai) iPaladin’s 2026 analysis goes further, naming the specific gap most family-office AI vendors leave unaddressed — where the data actually goes once a promise is made. [source: iPaladin →](https://www.ipaladin.com/architecture/what-every-family-office-ai-vendor-is-leaving-out) Most vendors respond with paperwork. We removed the surface the paperwork exists to cover.

- i **No third-party transit.** On-premise deployment means client documents and conversations never reach a third-party model provider. There is nothing upstream to retain them.

- ii **No other tenants.** There is no cross-tenant commingling because there are no other tenants. Your deployment serves one office: yours.

- iii **The contrast, plainly.** A "zero-retention API" and a "private instance" are policies — your data still transits someone else’s infrastructure, and the protection is a clause you trust. Ours is a wall your own network team can verify.

// the claim, in two sentences
The platform runs inside your walls. There is nothing of ours for your data to train.
— the one answer a policy can never match.
no training on your data no telemetry no shared tenant

own key vs. own infrastructure

## Your own key isn’t *your own infrastructure.*

Bring-your-own-key (BYOK) is the newest version of the same evasion. It lets a client hold the encryption key — but the data still transits and is processed on the vendor’s shared cloud platform. Encrypting data before it leaves the building is not the same as never sending it. On-premise means there is no vendor infrastructure in the path: nothing to encrypt for transit, because nothing transits.

the concrete surface

## What stays in the room — *all of it.*

"Nothing leaves" is easy to say and easy to blur. Here is the inventory, item by item, of what lives and remains inside your infrastructure.

##### 01 · in the room

documents
every statement, deed and agreement — stored and read inside your network

##### 02 · in the room

extractions
the structured data the platform reads out of them

##### 03 · in the room

chat history
every question asked and every answer given

##### 04 · in the room

embeddings
the vector index built from your documents

##### 05 · in the room

audit logs
the record of who saw what, and when — yours to hold

your network → your keys → your logs → your record

asked, plainly answered

## The questions *every office asks.*

### Does Super Chain train on our data?

No. Super Chain is deployed inside your own infrastructure, so your documents and conversations never reach a vendor-side model. There is no shared model to improve and no cross-tenant pool for your data to join — nothing leaves the room. The protection is architectural, not a clause in a policy.

### Where do our documents go?

Nowhere. Documents, extractions, chat history, embeddings and audit logs are all stored and processed inside your network — on the office’s own servers or a private tenancy your IT team controls. Super Chain keeps no copy, receives no telemetry, and has no vendor-side store for them to reach.

### Can our IT team audit it?

Yes. The platform runs where your team can watch it: your network, your keys, your logs. Every access and action is kept in an audit record that belongs to the office, and the absence of egress can be verified at your own firewall rather than taken on trust.

### What makes an AI architecture secure for a family office?

An architecture is secure for a family office when data never leaves the office’s control: the model runs on-premise, there are no other tenants to be commingled with, no vendor-side store to be breached, and the audit trail belongs to the office. Security by design, not by policy or perimeter add-ons. Mapped to the four questions a family office’s own security advisors typically ask of any AI vendor — data classification, identity and access control, vendor selection, and audit posture — Super Chain’s answer to each is architectural: per-org isolation and RBAC (classification and access), no vendor in the loop once deployed on-premise (vendor selection), and OpenTelemetry traces plus audit logging (audit posture).

### Is bring-your-own-key (BYOK) encryption the same as on-premise?

No. BYOK means the client controls the encryption key; the data still moves through and is processed on the vendor’s own cloud infrastructure. On-premise means the platform runs inside the family office’s own network — there is no vendor infrastructure the data ever has to reach.

### Where does the pilot run?

The 30-Day Office Memory Pilot runs on a dedicated, isolated deployment hosted by Super Chain, with an enterprise model API. When the office goes into production, the platform moves to its own infrastructure with a self-hosted model.

### How can a family office protect its data when using AI?

Start with custody, then access. As of October 2026, Super Chain’s approach is to run the AI on infrastructure the family office controls, limit each document to the people and roles cleared to see it, and keep an audit log the office holds. Add a written rule on which AI tools staff may use, and review every vendor’s data terms.

### What belongs in a cyber-safe AI roadmap for a family office?

As of October 2026, Super Chain suggests five steps. List where staff already use AI, and decide which documents may ever reach a model. Choose where the platform runs, set access by role and document, and keep an audit log the office can read. Each step is a decision for the principal and the COO, not for IT alone.

further reading — [the four tiers of "private AI" →](https://www.super-chain.tech/private-ai-for-family-offices)✦ [on-premise AI for wealth management →](https://www.super-chain.tech/on-premise-ai-for-wealth-management) · [case-file confidentiality for broker agencies →](https://www.super-chain.tech/ai-for-mortgage-brokers) · [how to protect family office data when using AI →](https://www.super-chain.tech/protect-family-office-data-when-using-ai) · [a cyber-safe AI roadmap for family offices →](https://www.super-chain.tech/cyber-safe-ai-roadmap-for-family-offices)

// the next step

## The 30-Day Office Memory Pilot

We load one vehicle’s documents onto a dedicated, isolated deployment hosted by Super Chain, with an enterprise model API. Your team asks real questions of them for 30 days, and each answer comes back cited to the page it came from. Production deployments run on-premise, on the office’s own infrastructure.
Choose a time below to scope the pilot with us. If you would rather see the platform first, [book a private demonstration](https://calendly.com/super-chain/catyung?utm_source=super-chain.tech&utm_campaign=private-demonstration&utm_content=trust).

[Open the scheduler →](https://calendly.com/super-chain/catyung?utm_source=super-chain.tech&utm_campaign=office-memory-pilot&utm_content=trust)
