# How to Protect Family Office Data When Using AI > Five controls, in order, for a family office using AI: custody, scope, access, audit and vendor terms, with who owns each and where a platform fits. Source: https://www.super-chain.tech/protect-family-office-data-when-using-ai Last modified: 2026-10-22 protecting family office data # How to protect family office data when using AI. Start with custody, then access. Decide which documents may reach an AI tool, run the tool on infrastructure the office controls, limit each document to the roles cleared to see it, and keep an audit log the office holds. A written rule for staff covers the tools nobody approved. custody · scope · access · audit · terms Written by [Cat Yung](https://www.super-cat.tech/cat-yung), Founder, Super Chain. Last updated 22 October 2026. where data leaks ## Where family office data *leaks into AI tools.* Most exposure comes from a document placed in a tool nobody approved. A statement goes into a free assistant, or a deed into a browser extension. Each hands the family’s affairs to an unreviewed service. In the 2026 RBC and Campden Wealth North America Family Office Report, published on 1 October 2026, 59 percent of 155 offices named cybersecurity and data breaches their top near-term operational concern. A year earlier it was 16 percent. (Source: [RBC Wealth Management, October 2026](https://www.rbcwealthmanagement.com/en-us/newsroom/2026-10-01/2026-rbc-and-campden-wealth-report-north-american-family-offices-shift-from-caution-to-optimism-amid-great-wealth-transfer-ai-acceleration).) five controls ## Five controls, *in order.* The order matters. Custody and scope decide most of the risk before any permission is set. - Custody.** Decide where the AI runs. A hosted tool processes data on the vendor’s servers. A platform on the office’s own private cloud or servers keeps that processing on infrastructure the office controls. - Scope.** List the documents that may never reach a hosted tool, by type: statements, deeds, side letters, anything naming the family. - Access.** Set permissions by role, down to the single document, so the analyst who reads one vehicle cannot query another. - Audit.** Keep a log of every question asked, held by the office, and read it on a schedule. - Terms.** Read the data terms of every AI vendor and model provider, date them, and read them again on renewal. who decides ## What the office decides, *and what a vendor can do.* The office decides which documents go in and who may see them. That decision sits with the principal or the COO, because it concerns the family’s information, with IT carrying out the work. A vendor can provide the deployment, the permission model and the log. The staff rule and the review stay with the office. For the questions staff ask most, read [is ChatGPT safe for family office data](https://www.super-chain.tech/is-chatgpt-safe-for-family-office-data). how super chain applies them ## How Super Chain applies *these controls.* Super Chain runs on the office’s own private cloud or servers, isolated from every other office. Access is set by role down to the single document, and each query is recorded in an audit log the office holds. Answers come from the office’s own documents, each cited to its page, and the office chooses the model the platform uses. The [trust and security page](https://www.super-chain.tech/trust) sets out the architecture. The [cyber-safe AI roadmap](https://www.super-chain.tech/cyber-safe-ai-roadmap-for-family-offices) sets out the order of decisions. who it is for ## Who this is for, *and who it isn’t.* This guide is for single and multi-family offices, and investment offices, whose staff use AI already or are about to, and whose documents carry family, tax and counterparty detail. It is not a full cyber programme. Email security, device management and penetration testing sit outside it, and an office needs those whether or not it uses AI. delivery ## How Super Chain is put in place, *and who does what.* - i **Scope.** The office names one owner and chooses one vehicle for the 30-Day Office Memory Pilot. Together we agree where the platform will run: the office’s private cloud or its own servers. - ii **Deploy.** Our team installs Super Chain on that infrastructure and configures roles, document permissions and the model the office chooses. - iii **Load and check.** The office loads that vehicle’s documents, not a tidied sample. We review the first answers with the owner against the cited pages. - iv **Use.** For 30 days the team asks real questions. Every query is recorded in an audit log the office holds. Super Chain is accountable for the platform and its configuration. The office remains accountable for which documents go in, who may see them, and the terms of the model provider it chooses. // what can go wrong - Access and security review take longer.** We list what is needed at the start and raise blockers the day they appear. - Some documents must stay in-house under the office’s policy.** We scope the deployment to that policy before any document moves. - A model or vendor changes its terms.** We check terms at selection and re-route to another model when they change. - Daily use differs from the pilot.** We measure on the office’s real questions and report the numbers as they are. // track record Super Chain is built by Super Cat Technology. The same team built a [fully on-premise retrieval system for a Big 4 firm](https://www.super-cat.tech/portfolio/enterprise-rag-big-4), where every answer returns with its source paragraph. “Most of the protection is decided before any software arrives. Which documents go in, who may see each one, and who signs off the model provider’s terms? My team builds the platform so those decisions become settings the office can check. Access is set down to the single document, and the office holds the audit log.” ## What Cat is seeing *in October 2026.* - October 2026 Offices ask who can see the trust deed before they ask which model reads it. Document-level access comes up in the first conversation now, not the last. - October 2026 Banks drew a line this month. Charles Schwab’s chief executive said on 16 October 2026 that outside AI agents may check balances and research but not move client money ([TNW, October 2026](https://thenextweb.com/news/schwab-blocks-ai-agents-meta-muse)). When a family office asks me about its own agent, I start from the same place: what it may read, and what it may never do without a person. - October 2026 Offices want the provider’s terms filed with the contract and dated, because terms change and the office carries that review. read next ## Related *reading.* Is ChatGPT safe for family office data? → When a hosted assistant is reasonable, and when the data needs a different home. A cyber-safe AI roadmap for family offices → Five decisions, who owns each, and where a platform fits. What “private AI” actually means → The four tiers of AI privacy, and the trade-offs of each. questions ## The questions offices ask, *plainly answered.* ### 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. ### Should a family office block public AI tools? A blanket ban tends to push use onto personal phones, where the office cannot see it. As of October 2026, Super Chain’s view is that a written rule works better. Name the approved tools and list the documents that may never go into a hosted tool. Then give staff a sanctioned place for work that needs the family’s documents. ### What should a family office ask an AI vendor about its data? Four questions cover most of it. Where does the software run, and who operates that infrastructure? Which model provider does it call, and on what terms? Can access be set per document? Is there an audit log the office holds? As of October 2026, Super Chain suggests asking for each answer in writing, dated, and filing it with the contract. ### Does running AI on-premise make family office data safe? It moves custody to the office; it does not finish the job. As of October 2026, Super Chain’s view is that on-premise or private-cloud deployment needs document-level permissions and an audit log beside it. Without those, anyone with access to the system can ask about anything, and nobody can show who asked. To test these controls on one vehicle’s documents, Start the 30-Day Office Memory Pilot → // the next step ## The 30-Day Office Memory Pilot We load one vehicle’s documents onto your infrastructure. Your team asks real questions of them for 30 days, and each answer comes back cited to the page it came from. 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=protect-family-office-data-when-using-ai). [Open the scheduler →](https://calendly.com/super-chain/catyung?utm_source=super-chain.tech&utm_campaign=office-memory-pilot&utm_content=protect-family-office-data-when-using-ai)