Legal
AI Policy
How we use AI, and how we do not.
Families ask us about artificial intelligence before they ask about almost anything else. Usually the question is a version of the same one: where does my information actually go, and who else sees it?
This page answers that. It covers how we use AI in our own work, how we approach it in the systems we build, and what we will not do.
1. The three commitments
One · We do not use your data to train any model. Not ours, not anyone else's. This is not a preference we might revisit — it is a condition of how we work.
Two · Your data is never submitted to a consumer AI service. We do not paste client information into ChatGPT, Claude, Gemini, Copilot chat or any other general-purpose assistant, whatever the convenience.
Three · Where AI touches your data, it does so inside a system you have agreed to. No new tool gets access to your information because someone found it useful. Our sub-processors are listed, and changes are notified.
2. Where AI sits in our own work
We run a business, and the business systems we use — for email, documents, project management and meeting notes — increasingly include AI features as standard. Pretending otherwise would be dishonest.
What we do about it:
- We select business or enterprise terms, not consumer accounts.
- Where a provider offers a setting under which customer content is not used to train their models, we choose it.
- We name every such system in our client agreements and in our Privacy Policy, so you can see what they are.
- We disable AI features where they are not needed for the work.
What we cannot do is warrant another company's internal practices. We can tell you which providers we use, on what terms, and what we have configured. We cannot audit their engineering. Anyone who tells you otherwise is overstating what they know.
3. Meetings and transcription
Where we record or transcribe a session, we tell everyone present before it starts and proceed only if everyone agrees. If anyone would rather we did not, we do not.
Recordings and transcripts are treated exactly as all other client information: confidential, encrypted, access-restricted, and deleted on the same terms as everything else.
Any client may tell us at any time not to record or transcribe their sessions, and not to process their information using any system with AI features. We will give effect to that without charge and without argument.
4. AI in the systems we build
When we build an operating system for a family, the architecture starts in the most private configuration and moves only if the family wants it to.
The default is no AI at all. The system runs on the family's own devices, reads a workbook the family owns, and makes no external connection. This is how we deliver unless asked otherwise.
If a family wants an AI layer, our position is settled:
- It runs on equipment the family controls, not on someone else's cloud.
- It works only against that family's own data.
- We never use Claude, ChatGPT, or any outside AI company on private family data.
- It can be switched off entirely at any time, and everything else keeps working.
- It is scoped, priced and agreed separately — never bundled into a build so that a family ends up with it by default.
Our sequencing is deliberate. Get the record right first. Introduce AI only where the return is obvious and the family is comfortable. A system nobody trusts is not a system.
Our Operating System Safety Guide sets out the options in plain terms, and we walk through them before any build begins.
5. How we use AI in what we publish
We write our own material. Where AI assists — with structure, research or a first draft — a person at Circle 26 has read, edited and stands behind every word we publish.
We do not publish anything we have not verified. Where we cite a figure, we say where it came from and as at what date.
We do not use AI to generate anything presented as a real person's words, and we do not use synthetic images of people.
6. What we will never do
- Use client data, or any personal data, to train an AI model
- Submit client information to a general-purpose consumer AI service
- Give an AI system authority to make or execute a decision affecting a family's affairs
- Deploy an AI capability for a client without their explicit agreement
- Pretend that a general-purpose assistant is a "private" system because we have a paid account
- Represent AI output as advice
7. Decisions stay with people
We do not make decisions about you by automated means alone.
Our assessment tools score answers against a published framework and return a profile. No conclusion is reached about anyone by a machine acting on its own, and nothing we produce is acted on without a person applying judgment to it.
Where we build an AI capability for a family, it answers questions about that family's own records. It does not decide anything, and it has no authority to act.
8. Changes
AI tools change quickly. Where our position or our tooling changes materially, we will update this page and tell clients directly.
If a provider we use changes its terms in a way that conflicts with the commitments above, we will change provider or change our configuration. The commitments come first.
9. Ask us
If you want to know exactly which systems touch your information, on what terms, and how they are configured, ask. We will tell you.
Circle Family Office Pte. Ltd. (company number 202526403M) 160 Robinson Road, #14-04 SBF Centre, Singapore legal@circle26.com