---
title: Is your family office's data being exposed through AI?
description: What happens to family office data in ChatGPT, Claude, Gemini and Copilot, by plan, from each provider's own policy pages, plus five tests for private AI.
canonical: https://privatesuperintelligence.si/blog/family-office-ai-data-exposure/
last-updated: 2026-10-16
---

# Is your family office's data being exposed through AI?

**Business plans from the major AI providers do not train on your content by default. The exposure sits in personal accounts, retention periods, outside processors and old sharing settings that AI can now read. Each provider's own policy, quoted below, shows where those lines are.**

We are the two founders of Private SuperIntelligence, and we wrote this for single-family office operators and principals, the multi-family offices and registered investment advisers (RIAs) that serve them, and venture firms holding founder and investor data. Every provider statement comes from that provider's own page, checked on 16 October 2026.

## Does this sound like your office?

- Someone on staff uses a personal ChatGPT, Claude or Gemini account for work. [1][7][11]
- Nobody has checked whether training is switched on in those accounts. [1][7]
- The office assumes a deleted chat is gone the same day. [3][8][11]
- Copilot or a connected app can open shared folders that were set up years before AI existed. [6][13]
- No one can name the companies, other than the AI provider, that handle what staff type. [5][14]

If two or more are true, most of the fixes are settings and decisions the office can make this month.

## Training and retention, by provider

Business plans from OpenAI, Anthropic, Google and Microsoft do not train on your content by default. Most personal plans can, depending on the person's settings; Microsoft's updated consumer Copilot app is the exception. Retention varies widely, and some providers keep certain chats long after the user deletes them.

| Provider and plan | Trains on your content by default? | How long it is kept |
|---|---|---|
| ChatGPT Free, Plus, Pro | Yes, unless the user opts out [1] | Deleted chats go within 30 days, with legal and security exceptions [3] |
| ChatGPT Business, Enterprise, API | No [1] | Enterprise admins set retention; deleted chats go within 30 days [2][4] |
| Claude Free, Pro, Max | Only if the user allows it, except chats flagged in a safety review [7] | Deleted chats go within 30 days. If training is allowed, kept chats can be held de-identified for up to 5 years [8] |
| Claude for Work and API | No, unless you send feedback or agree to it [9] | API inputs and outputs deleted within 30 days, with exceptions [10] |
| Gemini Apps (consumer) | Yes while Keep Activity is on [11] | 18 months by default; 72 hours with Keep Activity off; reviewed chats up to 3 years [11] |
| Google Workspace with Gemini | No, without your permission [12] | Admins control how long conversations are saved [12] |
| Microsoft Copilot (consumer app, updated 18 Aug 2026) | No: prompts, responses and files "aren't used to train foundation models" [15] | Kept in activity history until the user deletes it [15] |
| Microsoft 365 Copilot | No [13] | No fixed period published; kept as activity history under your Microsoft 365 terms [13] |

*Source: each provider's own policy pages. Plans and settings change, so re-check a row before you rely on it.*

Three details from those pages are easy to miss.

First, feedback can undo an opt-out. On OpenAI's personal plans, a thumbs up or thumbs down can make "the entire conversation associated with that feedback" eligible for training, even after the user opted out. Business plans do not work that way. [1]

Second, Google's consumer Gemini privacy page warns: "Please don't enter confidential information that you wouldn't want a reviewer to see or Google to use to improve our services." Chats that human reviewers looked at "are not deleted when you delete your activity." [11]

Third, legal holds override retention. In the New York Times lawsuit, OpenAI was ordered to preserve consumer ChatGPT and API content. That obligation ended on 26 September 2025, but OpenAI says it still stores limited historical user data from April to September 2025. ChatGPT Enterprise and Edu were excluded. [4]

![Training defaults by plan: personal plans from OpenAI, Anthropic and Google can train on conversations depending on the user's settings; Microsoft's updated consumer Copilot app does not; business plans from all four do not by default.](/images/articles/family-office-ai-data-exposure/training-defaults.svg)

*Source: OpenAI [1], Anthropic [7][9], Google [11][12], Microsoft [13][15].*

If we were auditing an office tomorrow, we would start with the personal account an analyst opened on a Sunday because it was faster. The business plan the office bought is the easier part.

![How long providers keep a chat: 72 hours at Google with activity off, 30 days for deleted chats at OpenAI and Anthropic, 18 months default activity at Google, up to 3 years for human-reviewed Gemini chats, up to 5 years de-identified in Anthropic training pipelines for kept chats when training is allowed.](/images/articles/family-office-ai-data-exposure/retention-clocks.svg)

*Source: OpenAI [3][4], Anthropic [8][10], Google [11], Microsoft [15].*

## Who else handles the data?

Each provider publishes a subprocessor list: the outside companies it uses to host, support and moderate the service.

OpenAI's list, last updated 14 October 2026, names Microsoft, CoreWeave, Oracle and Amazon Web Services among the cloud hosts for ChatGPT Business and Enterprise. It lists TaskUs, in the Philippines, for customer support, and for content moderation on the API and ChatGPT Business. [5]

Microsoft 365 Copilot now includes Anthropic as a subprocessor. Microsoft says it "enables Anthropic models on by default for most customers in commercial cloud," outside Europe and the UK. For a US office, that means a Copilot answer may come from a second AI company unless an administrator changes the setting. [14]

![The chain behind one prompt: the staff member, the AI provider, the cloud hosts, support and moderation contractors, and any model partner the platform routes to.](/images/articles/family-office-ai-data-exposure/data-chain.svg)

*Illustration. Sources: OpenAI sub-processor list [5]; Microsoft AI subprocessor documentation [14].*

In practice, an office that uses Copilot may now be using two AI companies.

## What can the AI open on its own?

Once AI is connected to email and shared drives, the office has to ask what the AI can read on a person's behalf, as well as what staff paste in.

Microsoft 365 Copilot "only surfaces organizational data to which individual users have at least view permissions," and Microsoft asks customers to use its permission tools so the right people see the right content. [13] Our reading: if the office's SharePoint was shared broadly in 2019, Copilot can surface those files to everyone who was given access then.

ChatGPT's apps are connectors to services such as Gmail, Google Drive and Slack. In ChatGPT Business, OpenAI says "many apps are enabled by default," while new Enterprise workspaces start with a selected set. An Enterprise organization can ask OpenAI Support to block staff from connecting the office's work email or drive accounts to a personal ChatGPT account. [6]

Agents, meaning AI that takes actions such as sending email or filing documents, also need a rule for which actions always wait for a person. In our view, sending money and sending email always should.

## What "private" should mean

A no-training promise answers one question. Before a family office gives AI the trusts, the deal room and the principal's inbox, it should have five more answered in writing:

1. **Where it runs.** On whose hardware, and in which location, is our information processed?
2. **Who runs it.** Who operates the system day to day, and who do they answer to?
3. **What reports home.** What does the system send back to the provider: content, logs, usage data, or nothing?
4. **Who sees what.** Can we set which staff, advisers and family members see which records, and which actions need a person?
5. **Where answers come from.** Can every answer be traced to the document behind it?

![Five tests for private AI: where it runs, who runs it, what reports home, who sees what, and where answers come from, each phrased as a question for any provider.](/images/articles/family-office-ai-data-exposure/five-tests.svg)

*Framework by Private SuperIntelligence.*

For any hosted service, the answer to the first two questions is the provider's data centers and the provider's staff. Hosted plans can still answer the others well: the pages cited above are clear on training, and Microsoft and OpenAI document permissions in detail.

## A five-step plan for this quarter

1. **List the accounts.** Ask every person which AI tools they used for work last month, personal accounts included. Make it safe to own up so the list is complete.
2. **Fix the defaults.** Switch off training on any personal account that touched office work, or move that work to a business plan. [1][7][11]
3. **Read three pages per provider.** Training, retention and the subprocessor list. Write the retention periods and subprocessor names into your AI policy.
4. **Tighten sharing before connecting anything.** Close old shared folders before Copilot or a connected app can read them. [13]
5. **Name the private information.** Decide what should never leave an environment the family controls: trusts, estate plans, health records, deal rooms, the principal's inbox. That information needs the five questions answered in writing.

RBC and Campden Wealth's release on their 2026 North America report says cybersecurity and data breaches became the top near-term operational concern, cited by 59% of respondents, up from 16% the year before. [16]

## Where Private SuperIntelligence fits

Private SuperIntelligence is a private AI system built for one family and run for it by Mitosis Labs, the company we co-founded. We built it so the family decides where it runs and what reports home, and the people running it answer to the family.

We build each deployment for one client, including the hardware, the choice of model provider and where the data is stored. The family decides which models process its information and what each provider can access. [20] Our published commitment is that a client deployment captures none of the client's data and sends no telemetry back to Mitosis Labs. [19] Deployments range from an isolated cloud environment to an air-gapped one with no connection to the outside internet, and the client holds the encryption keys. [18][20]

Day to day, it works across the office's own email, documents and records, answers with citations, and drafts within permissions the office sets; sending and payments wait for a person. [18] We set it up and run it, with a named contact, because Citi's May 2026 report found 57% of family offices cite lack of internal expertise as their biggest barrier to AI adoption. [17][18][19]

If your office is weighing a hosted enterprise plan instead, our [comparison of ChatGPT Enterprise and Private SuperIntelligence](/blog/chatgpt-enterprise-for-family-offices/) sets the two side by side.

## Frequently asked questions

### Does ChatGPT train on our family office's data?

On ChatGPT Business, Enterprise, Edu and the API, OpenAI says it does not train on your content by default. On personal Free, Plus and Pro accounts it may, unless the user opts out, and feedback can make a whole conversation eligible. [1]

### If we delete a chat, is it gone?

Not straight away. OpenAI and Anthropic remove deleted chats from their systems within 30 days, subject to legal and safety exceptions. Google keeps Gemini chats that human reviewers saw for up to three years, even after you delete your activity. Microsoft 365 Copilot keeps activity history under your Microsoft 365 terms. [3][8][11][13]

### Is Microsoft 365 Copilot private enough for a family office?

It does not train foundation models on your Microsoft 365 data, and it shows each user only what they can already open, so loose sharing settings carry over. Anthropic models are on by default for most commercial customers outside the EU, EFTA and UK. [13][14]

### What makes AI private rather than just enterprise-grade?

Written answers to five questions: where it runs, who runs it, what it reports home, who sees what, and whether answers cite their sources.

## Speak with the founders

Private SuperIntelligence takes a limited number of families at a time so every one gets the same standard of service. [20] Before a conversation, see where your own office stands: **[Take the AI data exposure check](/ai-data-exposure-check/?src=A2-blog)**, five questions that take under a minute.

Prakshal Jain (PJ), Co-founder and CTO, Private SuperIntelligence (by Mitosis Labs)

Alexander Morris, Co-founder and CEO, Private SuperIntelligence (by Mitosis Labs)

Disclosure: Prakshal Jain and Alexander Morris co-founded Mitosis Labs, which runs Private SuperIntelligence.


## Related reading

- [Private AI for family offices](/blog/private-ai-for-family-offices/)
- [Consolidated reporting across custodians, without the Excel export](/blog/family-office-consolidated-reporting/)
- [Tracking K-1s across 30 entities](/blog/k-1-tracking-across-entities/)
- [Capital call notices: catching every one, across every trust and LLC](/blog/capital-call-tracking/)
- [The family office technology stack](/blog/family-office-technology-stack/)
- [Family office operations, and where AI changes the staff math](/blog/family-office-operations-and-ai/)
- [What is private superintelligence?](/blog/what-is-private-superintelligence/)

## Sources

All provider pages were fetched and checked on 16 October 2026. Provider statements are quoted or paraphrased from each provider's own pages. Private SuperIntelligence service statements come from its published service pages.

1. [OpenAI: How your data is used to improve model performance](https://help.openai.com/en/articles/5722486-how-your-data-is-used-to-improve-model-performance)
2. [OpenAI: Data controls in ChatGPT](https://help.openai.com/en/articles/7730893-data-controls-faq)
3. [OpenAI: Deleting and archiving chats in ChatGPT](https://help.openai.com/en/articles/8809935-how-chat-retention-works-in-chatgpt)
4. [OpenAI: How we're responding to The New York Times' data demands](https://openai.com/index/response-to-nyt-data-demands/)
5. [OpenAI: Sub-processor list (last updated 14 October 2026)](https://openai.com/policies/sub-processor-list/)
6. [OpenAI: Connectors and apps in ChatGPT](https://help.openai.com/en/articles/11487775-connectors-in-chatgpt)
7. [Anthropic: Is my data used for model training? (consumer)](https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training)
8. [Anthropic: How long do you store my data? (consumer)](https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data)
9. [Anthropic: Is my data used for model training? (commercial)](https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training)
10. [Anthropic: How long do you store my organization's data?](https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data)
11. [Google: Gemini Apps Privacy Hub](https://support.google.com/gemini/answer/13594961)
12. [Google Workspace: Generative AI privacy hub](https://knowledge.workspace.google.com/admin/generative-ai/generative-ai-in-google-workspace-privacy-hub)
13. [Microsoft Learn: Data, privacy and security for Microsoft 365 Copilot](https://learn.microsoft.com/en-us/copilot/microsoft-365/microsoft-365-copilot-privacy)
14. [Microsoft Learn: AI providers operating as Microsoft subprocessors](https://learn.microsoft.com/en-us/copilot/microsoft-365/connect-to-ai-subprocessor)
15. [Microsoft: Copilot for individuals, your activity history](https://support.microsoft.com/en-us/privacy/microsoft-copilot/activity-history)
16. [RBC and Campden Wealth: release announcing the 2026 North America Family Office Report (September 2026)](https://www.prnewswire.com/news-releases/2026-rbc-and-campden-wealth-report-north-american-family-offices-shift-from-caution-to-optimism-amid-great-wealth-transfer-ai-acceleration-302892391.html)
17. [Citi Institute: AI in the Family Office (May 2026)](https://www.citigroup.com/global/insights/ai-in-the-family-office)
18. [Private SuperIntelligence: AI for family offices](https://privatesuperintelligence.si/family-offices/)
19. [Private SuperIntelligence: About](https://privatesuperintelligence.si/about/)
20. [What is private superintelligence?](https://privatesuperintelligence.si/blog/what-is-private-superintelligence/)