DEV Community: PJ (Prakshal Jain)

When family-office K-1s are due, and why they arrive in September

PJ (Prakshal Jain) — Thu, 05 Nov 2026 14:23:32 +0000

Disclosure: I co-founded Mitosis Labs, which runs Private SuperIntelligence. This is an adaptation of the original guide written for family offices. This is general information about IRS filing rules, not tax advice.

K-1 tracking is the job of knowing, for every fund, partnership and trust the family owns, which Schedule K-1 is due, when it is due, whether an estimate has arrived and whether the final has. Under IRS rules a partnership must give each partner a K-1 on or before the day its Form 1065 is required to be filed.

The calendar, then the chain

For tax year 2025 (returns filed in 2026), calendar-year partnerships file Form 1065 by March 15 (March 16 in 2026, because the 15th fell on a Sunday). Form 7004 is generally an automatic 6 months, so an extended partnership can send the K-1 in September. Trusts file Form 1041 by April 15, 2026, with a 5½-month Form 7004, and furnish K-1s to beneficiaries on or before the day the trust must file. Individuals who file Form 4868 have until October 15. An extension to file is not an extension to pay.

For 2025 returns the Form 1065 instructions set a $255 late-filing penalty per partner per month, up to 12 months, and a $340 penalty for each K-1 not furnished when due or furnished with wrong information. If the family office runs its own partnerships, those penalties land on entities you manage.

Outside funds send K-1s to holding LLCs and trusts. The family's own partnerships wait on those, then issue K-1s. Trusts wait next. Family members file Form 1040 last. One late fund K-1 can hold up a holding company, two trusts and four individual returns. That chain is an illustration, not a measured case. The Form 1041 instructions describe amended K-1s, which must be given to each beneficiary when an amended return changes their share.

What the tracker has to hold

Most offices start with a spreadsheet. Dedicated K-1 tracking software exists, and many offices already use one. Keep issuer and tax-document contact, receiving entity, form type (1065, 1041, 1120-S), last year's arrival date, estimate received, final received, amendment, Schedule K-3 expected, and downstream returns waiting on it.

AI can watch tax inboxes, match a K-1 to the right row, compare figures with last year and with the estimate, and draft chase mail for a person to send. Your CPA should still review every figure before it goes on a return. Citi Institute's May 2026 report quotes several CIOs who called the tax workload "highly manual and voluminous," and says AI could automate much of it, from data gathering to flagging potential issues. Citi also found that offices keep final decisions with experienced people.

Tax documents carry some of the family's most sensitive information. See the note on AI for CPAs and client confidentiality before you share K-1 data with any AI tool.

Private SuperIntelligence from Mitosis Labs is an AI concierge for family offices, enterprises and private clients. It works across the office's own email, documents and data, answers with citations, and acts within permissions you set, so it can draft every chase email while a person approves what goes out. The provider keeps no copy of the work. Dedicated tools exist, and many offices already use one. PSI is the option when the office wants an assistant that watches the inboxes and PDFs and leaves the filing calls with the controller and the CPA.

Speak with the founders

Private SuperIntelligence takes a limited number of families at a time so every one gets the same standard of service. Book a private call to talk through your office with Alex Morris.

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

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

Originally published at privatesuperintelligence.si.

Sources: IRS Form 1065 instructions 2025 https://www.irs.gov/instructions/i1065 ; IRS Form 1041 instructions 2025 https://www.irs.gov/instructions/i1041 ; IRS Form 7004 https://www.irs.gov/instructions/i7004 ; IRS extension page https://www.irs.gov/forms-pubs/extension-of-time-to-file-your-tax-return ; Citi Institute May 2026 https://www.citigroup.com/rcs/citigpa/storage/public/ai_in_the_family_office.pdf

Piece ID: K1-DEV

Claude, Copilot and Gemini: work plans do not train by default. A personal login is a different set of terms

PJ (Prakshal Jain) — Wed, 04 Nov 2026 14:24:52 +0000

Disclosure: I co-founded Mitosis Labs, which runs Private SuperIntelligence. This is an adaptation of the original guide written for family offices.

This is not a ranking of providers. A chief of staff who pastes a trust summary into a personal Gemini account and one who pastes it into Gemini through the family office's Workspace domain are under two different sets of terms. On Claude for Work and the Anthropic API, Microsoft Copilot signed in with a work Entra ID account, and Gemini in Google Workspace, all three providers say they do not train on your prompts by default.

Consumer versus commercial, in the providers' words

Since Anthropic updated its consumer terms on August 28, 2025, Claude Free, Pro and Max users choose whether chats may be used to train models. Allow it and Anthropic keeps that data for up to five years. Leave it off and 30-day retention continues, with exceptions for safety flags, thumbs up or down ratings (five years, de-linked from your user ID) and programs such as Trusted Tester. Incognito chats are not used to improve Claude. API inputs and outputs are deleted within 30 days. Flagged content can be kept up to two years. Zero data retention applies only to eligible APIs and Claude Code for Enterprise, not the Claude apps.

Microsoft states that work Copilot prompts, responses and Graph data "aren't used to train foundation models." Copilot answers from whatever the user can already open, so an over-shared SharePoint folder becomes an over-shared Copilot answer. Bing web queries sit outside the Data Protection Addendum. Anthropic models inside Copilot sit outside the EU Data Boundary. Copilot services have opted out of Azure OpenAI abuse monitoring, which includes human review. Personal Copilot trains unless you opt out, stores conversation activity 18 months by default, and some conversations get automated and human review for product improvement and digital safety. An opt-out of human review is not available when a Code of Conduct violation is suspected.

Google's Gemini Apps Privacy Hub says a subset of personal chats is reviewed by people, including service providers, and tells you not to enter anything you "wouldn't want a reviewer to see." Keep Activity off still leaves 72 hours of retention, and reviewed chats stay up to three years. Workspace treats prompts as customer data and says content is not human reviewed or used for training outside your domain without permission.

On ChatGPT Free, Go, Plus and Pro, training can be on unless you turn off "Improve the model for everyone." ChatGPT Business, Enterprise, Edu and the API do not train on your data by default. Those ChatGPT lines are from the canonical guide; OpenAI's enterprise-privacy page still returns HTTP 403.

Confirm everyone signs in with the work account. Ask what retention is set. Do not rate sensitive Claude chats. Fix file permissions before you switch Copilot or Gemini on across mail and documents. Each work plan still leaves the text on the provider's servers for a period.

Private SuperIntelligence from Mitosis Labs is an AI concierge for enterprises, family offices and private clients that works across your own email, documents and data and answers with citations. Each deployment is built for one client, and it captures none of that client's data. That is a different design from the work plans above, not a claim that those work plans are unsafe.

Speak with the founders

Private SuperIntelligence takes a limited number of families at a time so every one gets the same standard of service. Book a private call to talk through your office with Alex Morris.

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

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

Originally published at privatesuperintelligence.si.

Sources: Anthropic, August 28, 2025 https://www.anthropic.com/news/updates-to-our-consumer-terms ; Anthropic Privacy Center; Microsoft Learn Copilot enterprise data protection and privacy; Microsoft Support Copilot privacy FAQ; Gemini Apps Privacy Hub; Google Workspace Generative AI Privacy Hub.

Piece ID: ICG1-DEV

Citi found no perfect family-office system to buy. Offices still assemble five layers

PJ (Prakshal Jain) — Tue, 03 Nov 2026 14:21:08 +0000

Disclosure: I co-founded Mitosis Labs, which runs Private SuperIntelligence. This is an adaptation of the original guide written for family offices.

This is not a ranking of vendors. It is a map of five operating layers, the hand-offs between them, and what to ask before you switch on an AI feature.

Citi Institute's May 2026 report on AI in the family office found that the fintech market "currently offers no perfect single solution for family offices to buy," so most offices assemble a stack. The five layers are a portfolio and reporting system, accounting and entity records, bill pay and treasury, a document vault, and email, calendar and contacts. Citi also groups vendors into five categories (performance reporting; accounting and entity control; integrated platforms; private markets operations; governance and collaboration). Those are vendor buckets, not the five operating layers.

Addepar turns "multi-custodial data, illiquid alternatives and complex legal entities" into one source of truth. Masttro connects directly to "750+ custodians worldwide." Asset Vantage "unites general ledger and performance reporting." Eton lists bill pay and tax prep among the tasks EtonAI automates, inside a platform "Built by a Family Office for Family Offices." Asora is "Software for the modern family office" that automates data aggregation and customised reports.

The gaps, then the two kinds of AI

The portfolio system knows positions. Accounting knows the books. The vault holds the paper. A capital call notice still arrives by email. Citi found monthly statement consolidation still happens in Excel at many offices. Citi, citing Citi Wealth 2025, found that 57% of family offices name lack of internal expertise as the biggest barrier to technology adoption. Offices look for "software maturity, data readiness and comprehensive solutions that can be integrated easily." Most prefer to buy rather than build.

Addepar describes "native AI," Masttro offers "AI Powered Wealth Intelligence," and Eton calls itself "AI-powered." AI inside each product works on that product's data. An assistant that reads email, documents and reports together is a different job, with a person still approving what moves. Citi warns that some offices have access to AI through "the back door" via SaaS providers or devices they already use.

Before you switch on an AI feature: which model provider processes the data and on what retention terms; is the data used to train any model; who at the vendor can see family information; where is it stored and processed; can you turn the feature off per user or per entity. Those questions apply even when the feature is already on by default.

Private SuperIntelligence from Mitosis Labs is an AI concierge for family offices, enterprises and private clients. It works alongside the systems you already run, across the office's own email, documents and data, answers with citations, and acts within permissions you set. Each deployment is built for one client, and the provider keeps no copy of the work. That is the across-stack job, not a replacement for Addepar, Masttro, Eton, Asora or Asset Vantage.

Speak with the founders

Private SuperIntelligence takes a limited number of families at a time so every one gets the same standard of service. Book a private call to talk through your office with Alex Morris.

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

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

Originally published at privatesuperintelligence.si.

Sources: Citi Institute May 2026 https://www.citigroup.com/rcs/citigpa/storage/public/ai_in_the_family_office.pdf ; Addepar https://addepar.com/ ; Masttro https://masttro.com/ ; Eton https://www.eton-solutions.com/ ; Asora https://asora.com/ ; Asset Vantage https://www.assetvantage.com/

Piece ID: FOT1-DEV

A lean family office already costs millions. AI changes who does the prep

PJ (Prakshal Jain) — Mon, 02 Nov 2026 14:23:56 +0000

Disclosure: I co-founded Mitosis Labs, which runs Private SuperIntelligence. This is an adaptation of the original guide written for family offices.

This is not a promise that AI cuts the operating bill. It is a map of which work a tool can prepare, and which work still needs a person.

J.P. Morgan Private Bank surveyed 333 single-family offices in 30 countries, with an average net worth of $1.6 billion. Average annual operating costs by size: $0.9 million ($250 million or less), $1.7 million ($250 to $500 million), $3.3 million ($501 to $999 million), $6.6 million ($1 billion and above). The billion-dollar figure was $6.1 million in 2024. About 26% of costs go to external providers. 52% outsource legal, over 45% outsource trading and market execution, and 8 in 10 outsource some part of the investment portfolio. Cost is not a primary factor in those outsourcing decisions.

Citi describes those offices as lean teams that cannot scale technology the way institutions do. The calendar is still monthly statements, capital calls as they arrive, K-1s from March to October, weekly bill pay, and the principal's inbox every day. In a small office, one or two people hold the COO, controller and chief-of-staff jobs.

Where Citi says AI lands

Citi found family offices aiming AI at routine operations more than at investment returns. Most successful so far: summarizing documents, transcribing meetings, managing email and automating reports. 22% of family offices have automated operational tasks or use AI for investment analysis, up from 13% in 2024. Best results come from redefining roles, not from a savings promise. We give none. Time one cycle.

A cohort of offices led by tech-minded CIOs and principals aims for near-zero operational headcount, with a high bar of at least 80% efficiency gains before they adopt a tool. That last bar is a reminder, not a target to copy.

AI prepares vendor summaries, extracted figures, K-1 chase emails and draft replies. A person still keeps vendors, signs wires and decides what gets sent. Citi's interviewees describe drafts aligned with "the principal's intent and tone."

Why offices stall

Citi, citing the Citi Wealth 2025 Global Family Office Report: 57% name lack of internal expertise as the biggest barrier to technology adoption (34% awareness, 28% cybersecurity or privacy). Citi calls data security and privacy "the main concern for family offices and the primary barrier to broader AI adoption." Citi suggests an employee handbook, signed by staff, that requires approval for AI use cases. RBC and Campden 2026: 59% name cybersecurity and data breaches the top near-term operational concern, up from 16%. Half have an incomplete or non-existent succession plan.

We built Private SuperIntelligence as an AI concierge Mitosis Labs sets up and runs for one family office, across that office's own email, documents and data, with citations, within permissions you set. Delivery takes weeks.

Speak with the founders

Private SuperIntelligence takes a limited number of families at a time so every one gets the same standard of service. Book a private call to talk through your office with Alex Morris.

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

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

Originally published at privatesuperintelligence.si.

Sources: J.P. Morgan 2026 https://privatebank.jpmorgan.com/nam/en/insights/reports/2026-family-office-report ; Citi Institute May 2026 https://www.citigroup.com/rcs/citigpa/storage/public/ai_in_the_family_office.pdf ; RBC/Campden via PR Newswire 2026-09-29 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

Piece ID: FOO1-DEV

Family office cybersecurity now includes the AI tools on the perimeter

PJ (Prakshal Jain) — Fri, 30 Oct 2026 13:23:10 +0000

Disclosure: I co-founded Mitosis Labs, which runs Private SuperIntelligence. This is an adaptation of the original guide written for family offices.

This is not a reason to abandon AI. It is a reason to put AI vendors on the same cyber program you already run for custodians and IT providers.

Deloitte's Family Office Cybersecurity Report 2024 surveyed 354 single family offices between September and December 2023. 43% had a cyberattack in the previous 12 to 24 months globally, 57% in North America, and offices with more than US$1 billion AUM reported 62%, against 38% for smaller offices. Of attacked offices, 93% had received phishing emails. 31% had no cyber incident response plan. 68% had not adopted "know your vendor" protocols. Deloitte defines that as reviewing a vendor's security before you sign, for example by requesting security audit reports.

The 2026 RBC and Campden Wealth North America Family Office Report found that 59% name cybersecurity and data breaches the top near-term operational concern, up from 16% the year before. J.P. Morgan Private Bank's February 2026 report found that 32% cite cybersecurity as their greatest service need.

Three places AI widens the surface

IBM's 2025 Cost of a Data Breach Report, which studied 600 organizations, found that 16% of breaches involved attackers using AI tools, most often for phishing or deepfake impersonation. The FBI's IC3 received 22,364 complaints in 2025 that referenced AI, with losses of $893,346,472. Phishing, already 93% of Deloitte's attacked offices, got cheaper to write.

Staff paste family material into tools nobody approved. IBM uses "shadow AI" for that: one in five organizations reported a breach due to shadow AI; 63% of breached organizations had no AI governance policy or were still developing one. IBM did not break out family offices.

Assistants that read email can act. OWASP LLM01 is prompt injection, LLM02 is sensitive information disclosure, LLM06 is excessive agency. IBM found 13% of organizations reported breaches of AI models or applications, and 97% of those lacked proper AI access controls. Ask what the assistant can do without a person approving it.

Same Deloitte steps. Five vendor questions.

Identity by role. Vendor review. Training that includes cloned voices. An incident response plan that names who to call if an AI tool exposes family data. A maturity assessment that lists every AI tool in use, approved or not.

Ask whether a vendor trains on your data, how long it keeps it, where processing happens, what it can do without approval, and how you delete everything when you leave. Those five questions belong in the same vendor file as the custodian's security packet.

We built Private SuperIntelligence as an AI concierge for one family office, working across that office's own email, documents and data with citations, within permissions you set. A client deployment sends no telemetry back to Mitosis Labs.

Speak with the founders

Private SuperIntelligence takes a limited number of families at a time so every one gets the same standard of service. Book a private call to talk through your office with Alex Morris.

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

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

Originally published at privatesuperintelligence.si.

Sources: Deloitte 2024 https://www.deloitte.com/content/dam/assets-shared/docs/services/deloitte-private/2024/family-office-cybersecurity-report-2024.pdf ; RBC/Campden via PR Newswire, 2026-09-29 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 ; J.P. Morgan via PR Newswire, 2026-02-02 https://www.prnewswire.com/news-releases/jp-morgan-private-bank-releases-2026-global-family-office-report-302676012.html ; IBM Newsroom, 2025-07-30 https://newsroom.ibm.com/2025-07-30-ibm-report-13-of-organizations-reported-breaches-of-ai-models-or-applications,-97-of-which-reported-lacking-proper-ai-access-controls ; IC3 2025 https://www.ic3.gov/AnnualReport/Reports/2025_IC3Report.pdf ; OWASP LLM Top 10 2025 https://genai.owasp.org/llm-top-10/

Piece ID: FOC1-DEV

Family office reporting still ends in Excel. The PDFs are why.

PJ (Prakshal Jain) — Thu, 29 Oct 2026 13:32:43 +0000

Disclosure: I co-founded Mitosis Labs, which runs Private SuperIntelligence. This is an adaptation of the original guide written for family offices.

Citi Institute's May 2026 report on AI in the family office found that monthly reports from advisers are manually consolidated into Excel due to privacy concerns. The 2026 RBC and Campden Wealth North America Family Office Report found that 73% of offices believe investment reporting is too manual, a top concern for three years running. The ingestion problem is documents, not a missing feed.

Custodian APIs do not eat the PDF pile

RBC and Campden found that 86% of North American family offices invest in private markets, and that direct investments now average 45% of that book, ahead of funds at 36%. Capital account statements, valuation letters, K-1s and ownership changes arrive as PDFs and email. Portfolio software such as Addepar, Masttro or Asora aggregates custodian data and handles complex entity structures. Addepar describes turning "multi-custodial data, illiquid alternatives and complex legal entities" into one source of truth. The inputs still have to be read first.

Citi names data privacy as non-negotiable and reports that offices pass over AI tools that cannot guarantee data security. Interviewees worry about storing sensitive financial data and about third-party vendor access. Citi also found many offices lack confidence in AI-produced reports, and calls this "a major roadblock to full AI automation." If you would re-check every number anyway, Excel on the office's machines is easier to defend.

Four jobs for a reader, not a new system of record

Citi's interviewees describe ingesting data from custodian APIs, extracting information from tax forms, and processing insurance documents into consolidated views. In a reporting cycle that is:

Collect from inboxes and portals. Log gaps.
Extract figures with a citation to the page.
Reconcile against the portfolio system or ledger. Flag a call booked to the wrong entity, a distribution that does not match the notice, a statement that disagrees with last quarter's close.
Draft commentary a person validates. Citi found the most sophisticated users treat AI output as a draft, starting with a full review and moving to spot checks as confidence builds.

Ask every vendor where data lives and whether you can choose the location, whether it or its model providers train on yours, who can see family information, whether every number cites its source, whether it works with software you already run, and how you delete it at the end of the contract. Replacing the system of record is usually the wrong first project.

We built Private SuperIntelligence as an AI concierge for one family office, working across that office's own email, documents and data with citations, within permissions you set, alongside the portfolio software you already run. The provider keeps no copy of the work and trains nothing on it.

Speak with the founders

Private SuperIntelligence takes a limited number of families at a time so every one gets the same standard of service. Book a private call to talk through your office with Alex Morris.

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

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

Originally published at privatesuperintelligence.si.

Sources: Citi Institute, May 2026 https://www.citigroup.com/rcs/citigpa/storage/public/ai_in_the_family_office.pdf ; RBC/Campden via PR Newswire, 2026-09-29 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 ; Addepar, Masttro and Asora home pages, accessed 2026-10-15. Addepar fragment rechecked 2026-10-29.

Piece ID: FCR1-DEV

A one-page AI acceptable-use policy you can actually get signed

PJ (Prakshal Jain) — Wed, 28 Oct 2026 13:38:17 +0000

Disclosure: I co-founded Mitosis Labs, which runs Private SuperIntelligence. This is an adaptation of the original guide written for family offices.

IBM's 2025 Cost of a Data Breach Report found that 63% of breached organizations either had no AI governance policy or were still developing one, and only 37% had policies to manage AI or detect shadow AI. Organizations with high levels of shadow AI saw breach costs $670,000 higher on average. Most small professional offices need a page people will sign.

The access problem is account type

On ChatGPT's personal Free, Plus and Pro plans, OpenAI may use chats for training unless the user turns off "Improve the model for everyone". Business, Enterprise, Edu and the API don't train by default. A staff member who pastes a client memo into a personal account has chosen the terms, not the firm.

Eight clauses that fit on one page

Scope. Every employee, contractor and adviser, any device.
Approved tools. Name each tool and the account type. Personal AI accounts are not approved, even with training off.
Never list, in nouns. Tax returns, K-1s, trust and estate documents, account numbers, wire instructions, passwords, health information, travel plans, home addresses, security arrangements, passports, anything a client asked to keep private.
Agency. Tools may read, summarize and draft. A named person approves before send, share, record change or payment. No AI tool releases a wire. OWASP's 2025 LLM Top 10 calls the failure mode "excessive agency".
Output check. The human who sends it owns every figure, date and name.
New tools. Ask the owner first. Five vendor questions: training, retention, processing location, permissions, exit.
Incidents. Same-day report. No discipline for reporting quickly.
Outside advisers. Ask which tools they use on your matters.

Start the approved list by asking, without blame, what people already tried in the last three months. Two tools they use beat a ban they work around.

Mapping to NIST without adopting the whole RMF

NIST's AI RMF 1.0 (NIST AI 100-1, January 26, 2023) is voluntary. Four functions: Govern, Map, Measure, Manage. Owner and sign-off are Govern. Approved tools and the never list are Map. Output checks and vendor review are Measure. Approval limits and incident reporting are Manage. NIST added a Generative AI Profile (NIST AI 600-1) in July 2024. That mapping gives an investment committee or insurer a short, standard answer about the written policy.

Keep a dated register of approved tools and decisions. Read the policy aloud with staff. Review every six months and sooner when a tool can act on the family's behalf. This is general information, not legal advice.

We built Private SuperIntelligence as an AI concierge for one family office, working across that family's own email, documents and data with citations, within permissions the office sets, so clause 4 is a configuration, not a hope.

Speak with the founders

Private SuperIntelligence takes a limited number of families at a time so every one gets the same standard of service. Book a private call to talk through your office with Alex Morris.

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

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

Originally published at privatesuperintelligence.si.

Sources: NIST AI RMF https://www.nist.gov/itl/ai-risk-management-framework ; IBM Newsroom, July 30, 2025 https://newsroom.ibm.com/2025-07-30-ibm-report-13-of-organizations-reported-breaches-of-ai-models-or-applications,-97-of-which-reported-lacking-proper-ai-access-controls ; OWASP Top 10 for LLM Applications 2025 https://genai.owasp.org/llm-top-10/ ; OpenAI Help Center on model training and ChatGPT Business.

Piece ID: FOP1-DEV

Your threat model has a video call in it now: deepfake payment fraud and controls that don't need detection

PJ (Prakshal Jain) — Tue, 27 Oct 2026 13:45:41 +0000

Disclosure: I co-founded Mitosis Labs, which runs Private SuperIntelligence. This is an adaptation of the original guide, written for family offices.

If you build or secure anything that moves money, here is a definition worth reading closely. The FBI's Internet Crime Complaint Center describes business email compromise as fraud carried out "by compromising email accounts and other forms of communication such as phone numbers and virtual meeting applications."

Virtual meeting applications. The video call is now in scope as an attack channel, not a verification step. The FBI's December 2024 public service announcement is more specific: criminals "generate videos for real time video chats with alleged company executives."

The case that proves the point

CNN reported in May 2024, citing Hong Kong police, that a finance worker in Arup's Hong Kong office received an email that appeared to come from the UK office, asking for a secret transaction. He suspected phishing, which was the right call. Then he joined a video call with people who looked and sounded like the CFO and colleagues he knew. All were deepfakes. He sent HK$200 million (about $25.6 million) across 15 transactions.

Arup told CNN that "fake voices and images were used" and "none of our internal systems were compromised." Rob Greig, Arup's global chief information officer, said the company faces "invoice fraud, phishing scams, WhatsApp voice spoofing, and deepfakes," and that their number and sophistication "has been rising sharply in recent months."

Read that as an incident report. No credential theft, no malware, no lateral movement. The human detection layer worked once (he doubted the email) and was then defeated by a richer signal. Any control that depends on a person judging authenticity has the same failure mode.

Scale, from the source data

From the IC3 2025 annual report:

Measure (IC3, 2025)	Figure
BEC complaints	24,768 (21,442 in 2024)
BEC losses	$3,046,598,558
BEC losses where the complaint referenced AI	over $30 million
All complaints referencing AI, any crime type	22,364, losses $893,346,472
Distress scams using cloned voices of loved ones	over $5 million

FinCEN's alert FIN-2024-Alert004 (November 2024) says criminals "have reportedly used GenAI tools to target companies by impersonating an executive or other trusted employee and then instructing victims to transfer large sums."

Family offices sit inside this target group. Deloitte's Family Office Cybersecurity Report 2024 found that 93% of family offices that had been attacked had received phishing emails, and nearly a quarter had faced social engineering.

Controls that are independent of detection

The design principle: authorization must not depend on the authenticity of the requesting channel. Each of these holds against a perfect fake.

Out-of-band callback to a pre-registered number. The verifier initiates contact on a number already in your records. Never a number, link or invite carried in the request. The FBI gives the same advice.
Payee changes are new payees. Any change to bank details triggers the callback and a holding period, regardless of sender.
Two-person release. One prepares, a second approves and confirms the callback occurred.
Shared secret for humans. The FBI suggests a family code word. Keep it off email and messaging apps.
Secrecy as a hard stop. A request to hide a payment from the controller is policy-defined grounds to halt.
Video is not verification. Faces and voices can be generated. Verification happens on a separate channel you started.
Phishing-resistant MFA. FinCEN lists MFA, including phishing-resistant MFA, as a best practice against deepfake identities.

FinCEN's red flags are written for banks, but two port well: the counterparty asks to switch channels mid-verification, or blames "technical glitches" for a camera or line that keeps failing. Both should terminate the session and trigger a callback.

Five questions before the next wire

Do we call back on a number already in our records, never one in the request?
Is a change of bank details treated as a new payee with a callback and a holding period?
Do two people release every wire, with the approver confirming the callback happened?
Have we agreed a family code word in person, and is it kept off email and messaging apps?
Is a request for secrecy, or a video call used as verification, written as a stop?

Incident response

The IC3 says "time is of the essence." Contact the bank and request a recall, then file at ic3.gov with full transaction details. In 2025 the IC3's Recovery Asset Team worked 3,900 incidents through its Financial Fraud Kill Chain and froze $679,013,183, a 58% success rate.

Where an AI assistant belongs in this

If you are wiring an assistant into a payments workflow, scope it to read, compare and draft: match a capital call to its commitment and the bank details on file, flag requests that arrive with new instructions, draft the callback note. Do not grant it permission to release payments. Private SuperIntelligence works that way: it acts within permissions the office sets, so a payment can stay a draft until a person approves it.

Speak with the founders

Private SuperIntelligence takes a limited number of families at a time so every one gets the same standard of service. Book a private call (ref DFW1-DEV) to talk through your office with Alex Morris.

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

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

Sources (snapshot 2026-10-24, primary fetches 2026-10-17; recheck on publish day 2026-10-27): 2025 IC3 Annual Report; FBI IC3 PSA, Dec 3, 2024; FinCEN FIN-2024-Alert004; CNN, May 16, 2024; Deloitte Family Office Cybersecurity Report 2024.

Originally published at privatesuperintelligence.si.

Copilot inherits your permissions: an oversharing audit before you turn it on

PJ (Prakshal Jain) — Mon, 26 Oct 2026 13:53:13 +0000

Disclosure: I co-founded Mitosis Labs, which runs Private SuperIntelligence. This is an adaptation of the original guide written for family offices.

If you run IT for a small organization on Microsoft 365, Copilot changes one thing about your permissions model: it removes the effort it used to take to find things. This post walks through Microsoft's own pre-rollout guidance with the small-tenant case in mind.

The access model in one sentence

From Microsoft Learn: "Microsoft Copilot only surfaces organizational data to which individual users have at least view permissions." The semantic index honors each user's access boundary.

The privacy side is well covered. Microsoft states that prompts, responses and data accessed through Microsoft Graph aren't used to train foundation LLMs, that Copilot services have opted out of Azure OpenAI abuse monitoring with human review, and that Purview sensitivity labels and their usage rights are honored. Microsoft also describes classifiers that help block jailbreak and cross-prompt injection attacks.

So the risk surface is your existing ACLs.

What Copilot can reach

Source	Reach
Email, chats, documents, meetings, calendar, contacts	Anything in Microsoft Graph the user can access
SharePoint and OneDrive	SharePoint permissions, including sharing links
Teams shared channels	Includes access granted to external people
Graph connectors	Third-party data, if the user has permission
Agents	A query built from the prompt, activity history and the user's M365 data, sent to admin-allowed agents
Web search	A generated query sent to Bing

Microsoft's three-step blueprint

1. Remediate oversharing. Use Purview Data Security Posture Management and the SharePoint Advanced Management (SAM) Content Management Assessment to find overshared, ownerless or inactive sites. Apply Restricted Content Discovery to keep sensitive sites out of Copilot while you work. Run site access reviews, remove groups like "Everyone except external users", and fix broken permission inheritance.

2. Set guardrails. Restricted Access Control on business-critical sites, restrict Anyone links and company-wide sharing at tenant level, require sensitivity labels, and add Purview DLP policies so labeled files or sensitive prompts are excluded from Copilot.

3. Meet regulations. Set retention for audit logs and Copilot interactions; use Purview eDiscovery to search and preserve Copilot content.

Licensing notes from Microsoft: SAM is included with Copilot licenses; the blueprint needs at least foundational Purview features in Microsoft 365 E3, with some optimized features in E5 or E7. Viewing the Copilot security dashboard needs Global Reader; changes need AI Administrator.

Two things people miss

Activity history is a record. Prompts and responses (with citations) are stored, searchable via Content search or Purview, and subject to your retention policies. Treat them like email for legal hold.
Processing location. Microsoft routes to the closest regional data centers but may use other regions under high load. EU traffic stays within the EU Data Boundary; outside the EU, queries may be processed in the US, the EU or elsewhere. If admins enable Anthropic models inside Copilot, Microsoft notes they are currently excluded from the EU Data Boundary.

Five questions before you switch it on

Who holds Global Reader and AI Administrator in our tenant?
Which SharePoint sites hold trust, estate or principal documents, and who can open them today?
Have we run an oversharing assessment and turned on Restricted Content Discovery for sensitive sites?
How long do we keep Copilot interactions, and who decided?
Which connectors and agents are allowed, and have we read their terms?

Why this matters more in a family office

A family office tenant can hold trust, estate and principal documents, alongside an old sharing link to an outside accountant or a site for a sold operating company that nobody owns any more. Before Copilot, a junior analyst would have had to go looking for them; with Copilot, a question about "the 2019 trust distributions" can surface them. The audit above is what keeps that answer inside the right audience. We built Private SuperIntelligence as an AI concierge for one family, working across the family's own email, documents and data with citations, within permissions the family sets.

Speak with the founders

Private SuperIntelligence takes a limited number of families at a time so every one gets the same standard of service. Book a private call (ref CPL1-DEV) to talk through your office with Alex Morris.

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

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

Originally published at privatesuperintelligence.si.

Sources (Microsoft Learn, snapshot 2026-10-24): Data, Privacy, and Security for Microsoft Copilot https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy; Security for Microsoft Copilot https://learn.microsoft.com/en-us/microsoft-365/copilot/security-microsoft-365-copilot; Secure and governed data foundation for Microsoft Copilot (deployment blueprint) https://learn.microsoft.com/en-us/microsoft-365/copilot/secure-govern-copilot-foundational-deployment-guidance; Configure a secure and governed foundation https://learn.microsoft.com/en-us/microsoft-365/copilot/configure-secure-governed-data-foundation-microsoft-365-copilot.

Capital call notices: catching every one, across every trust and LLC

PJ (Prakshal Jain) — Fri, 23 Oct 2026 13:24:32 +0000

Disclosure: I co-founded Mitosis Labs, which runs Private SuperIntelligence. This is an adaptation of the original guide.

If you build internal tools for a family office, capital calls are a good example of a workflow where AI should do the reading and a person should keep the money. This post sketches the tracker: what to store, how notices get in, what the model is allowed to do, and where the hard stop sits.

The input problem

A capital call notice is a fund manager's request for part of the money a limited partner committed. Notices arrive by email and through fund portals, addressed to different people and different entities. Search for capital call management software and most results are built for the fund side, the GP and its administrator, so offices often build the LP-side tracker themselves, many of them in Excel. Some family office systems now cover it: Masttro, for example, lists capital calls and reconciliation among its features.

Formats vary. ILPA released an updated Capital Call and Distribution Template in 2025 with a standalone LP unfunded commitment section, but says funds launched on or after Q1 2027 should use it (or from Q1 2026 for funds adopting its new performance template). Your parser has to handle each manager's layout for years.

A schema that holds up

Field	Why it exists
fund, receiving_entity	One fund can have several family investors (two trusts, one LLC)
notice_date, due_date	Funding deadline
amount_investments, amount_fees, amount_expenses	Lets you check the call against the LPA
unfunded_before, unfunded_after	Feeds the liquidity plan and consolidated report
wire_instructions, matches_file	First fraud check
funding_account, approver	Who released the money
wire_confirmation	Closes the loop with the fund

Key the record on fund plus receiving entity, not fund alone. "Two trusts each hold an interest, and the team pays one and misses the other" is one of the failure modes the original guide lists.

The extraction step

ILPA's best practices tell you what a well-formed notice contains: balances before and after the call (unfunded commitment, cumulative contributions and distributions), an itemized list of the investments funded, management fee details, LPA references, payment instructions and a due date. That list doubles as your extraction target and your validation checklist.

What the AI layer does:

watch every inbox and portal alert and classify capital call and distribution notices
extract each field, citing the notice page it came from
recompute unfunded commitment and flag any disagreement with the GP's stated balance
diff the wire instructions against the instructions on file and flag any change
draft the funding request for the approver with the notice attached

Citations matter here. A reviewer should be able to click from the extracted due date to the line in the PDF it came from.

The gate that stays human

The FBI's Internet Crime Complaint Center reported $3,046,598,558 in business email compromise losses in 2025, and defines the scheme as targeting businesses or individuals who work with suppliers or businesses that regularly perform wire transfers. A capital call is a request of exactly that kind. So the design has three permanent human steps:

approval of every wire
a phone call, to a number from your own records, confirming any changed bank details
the choice of funding account

Separate the person who logs the call from the person who releases the wire. No single message, email or voice should be able to move money. Citi Institute's May 2026 report on AI in the family office notes that offices must be mindful of risks in treasury and finance functions, given fraud and cyber threats, and that final decisions stay with experienced people.

Where we fit

Private SuperIntelligence from Mitosis Labs is an AI concierge for family offices. It works across the office's own email, documents and data, answers with citations, and acts within permissions the office sets, so it can log every notice and prepare each funding request while the release of money stays with your team.

Private SuperIntelligence takes a limited number of families at a time so every one gets the same standard of service. Book a private call (ref CCT1-DEV) to talk through your office with Alex Morris.

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

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

Originally published at privatesuperintelligence.si.

Sources

Capital Call and Distribution Notice Best Practices, Version 1.1 (ILPA, September 2016): https://ilpa.org/wp-content/uploads/2018/03/ILPA-Best-Practices-Capital-Calls-Distribution-Notices_Version-1.1.pdf
Capital Call and Distribution Template (ILPA, accessed 2026-10-15): https://ilpa.org/resources-tools/resource-library/capital-call-distribution-template/
ILPA Updates Capital Call and Distribution Template (Katten, 2025-11-26): https://katten.com/ilpa-updates-capital-call-distribution-template-implications-for-gps-and-lps
2026 RBC and Campden Wealth North America Family Office Report (PR Newswire, 2026-09-29): 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
2025 IC3 Annual Report (FBI Internet Crime Complaint Center): https://www.ic3.gov/AnnualReport/Reports/2025_IC3Report.pdf
AI in the Family Office (Citi Institute, May 2026): https://www.citigroup.com/rcs/citigpa/storage/public/ai_in_the_family_office.pdf
Masttro home page (Masttro, accessed 2026-10-15): https://masttro.com/

Mapping a VC firm's AI stack by data flow: connectors, MCP and training defaults

PJ (Prakshal Jain) — Thu, 22 Oct 2026 17:39:00 +0000

Disclosure: I co-founded Mitosis Labs, which runs Private SuperIntelligence. This is an adaptation of the original guide.

If you are the person at a venture firm who wires up the tools, "which AI should we use?" is really a data-flow question. Each tool reads some slice of founder and LP data, through some connector, under some training default. Draw that for each candidate before comparing features.

Everything I say about vendors below comes from their own pages, accessed 2026-10-15. Sources are at the end.

The data-flow view

Group	Tools	What the tool reads	Training default the vendor states
General assistants	ChatGPT Business/Enterprise, Claude for Financial Services, Microsoft Copilot	What users paste, upload or connect; for Copilot, Microsoft Graph content the user can already access	Each says business data is not used for training by default (Microsoft: not used to train foundation LLMs)
Sourcing and relationships	Harmonic, Affinity	Vendor's company database plus what you sync; for Affinity, your pipeline, meetings and email	Not covered in the original guide; ask
Documents and diligence	Hebbia, V7 Go	The documents and data you load	Hebbia: never used to train its or providers' models. V7: "No training on your data"
Portfolio and LP reporting	Visible, Standard Metrics	Portfolio data, founder updates and documents	Not covered in the original guide; ask

Connectors are where the scope expands

Three details matter more than the feature lists.

Inbox and calendar sync. Harmonic offers network syncing from inboxes, calendars and LinkedIn. Affinity's agents "prep your meetings, capture conversations, and write updates back to your pipeline." Once a tool syncs mail, it reads relationship data, not just company data. Affinity's security page lists SOC 2 Type II and ISO 27001 and encryption at rest and in transit, but did not cover model training or which model providers process data, so put that question in writing.

MCP servers. Visible connects AI workflows through MCP and Standard Metrics ships an MCP server for third-party LLMs. That is convenient, and it means portfolio data can flow into whatever model the user connects. Your data boundary becomes the union of both vendors' terms.

Permission inheritance. Microsoft says Copilot works over Graph content "that you have permission to access." Copilot's reach is your SharePoint and Teams ACLs. Fix oversharing before rollout. Microsoft also notes that models provided by Anthropic as a subprocessor are currently excluded from the EU Data Boundary.

Certifications as stated

OpenAI: SOC 2 Type 2, AES-256 at rest; Enterprise admins control retention, deleted conversations removed within 30 days unless law requires otherwise.
Harmonic: ISO 27001, SOC 2 Type II.
Hebbia: SOC 2 Type 2, ISO 27001:2022, ISO/IEC 42001:2023, US and EU regional processing, dedicated tenant "if needed".
V7 Go: "Encrypted end-to-end", access controls, audit logs.
Standard Metrics: "We are SOC 2 and GDPR certified."

A five-line vendor checklist

Does it train on our data, and does the answer cover the model providers it uses?
How long does it keep documents, prompts and outputs, and who on its staff can read them?
Does every figure it produces cite a source document?
Can it act outside the firm (email, CRM writes, sharing), and can we require approval?
Can we describe it, function by function, in an LP's DDQ?

Visible's AI Sentiment Report found tool overload (49%) almost as common a blocker as security concerns (51%), so the checklist is also a filter for fewer tools.

Where Private SuperIntelligence fits

Private SuperIntelligence is an AI concierge that works across a firm's own email, documents and data, answers with citations and acts within permissions the firm sets. Each deployment is built for one client, from the hardware to the choice of model provider to where the data is stored. It is not a CRM or startup database and runs beside them.

Private SuperIntelligence takes a limited number of families at a time so every one gets the same standard of service. Book a private call (ref BVC1-DEV) to talk through your office with Alex Morris.

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

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

Originally published at privatesuperintelligence.si. Vendor pages accessed 2026-10-15.

Sources

Enterprise privacy at OpenAI (OpenAI), accessed 2026-10-15
Data controls in ChatGPT (OpenAI Help Center), accessed 2026-10-15
Claude for Financial Services (Anthropic), accessed 2026-10-15
Data, Privacy, and Security for Microsoft Copilot (Microsoft Learn), accessed 2026-10-15
Harmonic (Harmonic), accessed 2026-10-15
Affinity (Affinity), accessed 2026-10-15
Enterprise-grade security (Affinity), accessed 2026-10-15
Hebbia (Hebbia), accessed 2026-10-15
Security (Hebbia), accessed 2026-10-15
V7 Go (V7), accessed 2026-10-15
Visible (Visible), accessed 2026-10-15
AI Sentiment Report Vol. 1 (Visible, 2026), accessed 2026-10-15
Standard Metrics (Standard Metrics), accessed 2026-10-15

Shortlisting AI for a family office: map data reach, permissions and training defaults first

PJ (Prakshal Jain) — Wed, 21 Oct 2026 13:12:29 +0000

Disclosure: I co-founded Mitosis Labs, which runs Private SuperIntelligence. This is an adaptation of the original guide.

If you are the technical person at a family office, "which AI tool should we use?" usually lands on your desk as a product question. I find it easier to treat it as a data-access question. Every tool on this list reads some set of the office's information, under some permission model, with some default about training and retention. Draw those three things for each candidate and the shortlist gets much shorter.

Everything I say about vendors below comes from their own pages, or for Aurora from a Fortune report, accessed 2026-10-15. Sources are at the end.

Boundary 1: what can the tool reach?

A family office holds a portfolio, the principal's email, entity and trust documents, the household and travel. The tools on offer split into four groups by how much of that they reach.

Category	Examples	What it reads	Where its scope ends
Enterprise AI seats	ChatGPT Business or Enterprise, Microsoft Copilot, Claude Enterprise	What the office connects	The office sets it up and governs it under the provider's terms
Platforms adding AI	Addepar, Masttro, Asora, Eton, Aleta	The portfolio's system of record	Answers come from the data in the platform
Private chat apps	Proton Lumo, Tinfoil, Venice	What one person types	No link to the office's own information
Concierge services	Knightsbridge Circle, Aurora	Lifestyle and travel requests	Outside the office's records and entities

Microsoft says Copilot works across email, chats, documents and meetings in Microsoft Graph. Anthropic lists Claude Enterprise connectors to Gmail, Google Drive, Slack and Microsoft 365. So with enterprise seats, reach is a configuration decision your office owns.

The platforms reach the other way: deep into portfolio data, and stop there. Masttro's DocAI ingests PDFs for liquid and illiquid assets and handles capital calls, distributions and valuation statements. Eton says EtonAI does document extraction across 250+ document types, with natural language querying and a knowledge graph. Aleta reads capital calls, K-1s and fund statements. Addepar, in a March 2026 post, says Addison "draws on Addepar's unified data model to answer questions, surface relevant guidance and direct users to the right workflows." Each assistant answers from what sits in its platform, so the principal's inbox, the household's calendar, deal email and travel stay outside it.

Boundary 2: MCP is moving where the answer is produced

Two of the platforms now expose portfolio data to outside assistants through the Model Context Protocol.

The Asora AI Connector lets you query Asora wealth data from Claude, ChatGPT, Gemini, Grok or any MCP-capable assistant. Access is read-only, scoped to existing permissions, and every query is audit-logged.
Aleta supports MCP so AI agents can query the family's wealth data in plain language.

This is good design, and Asora is clear about the trade: it says it never uses your prompts or data for training, and notes that once an answer reaches your assistant, that provider's terms apply. For your data-flow diagram, that means the trust boundary now runs through the assistant you connect, and that assistant's retention and training settings matter as much as the platform's.

Boundary 3: permissions are inherited, including the old ones

Copilot is the clearest example. Copilot shows each user any content that user has at least view permission for. If a trust folder was shared with the whole company years ago, Copilot can surface it to anyone in the company. Copilot is respecting the permissions you already have, so old sharing settings on trust and entity folders matter. Review them before rollout.

ChatGPT Enterprise and Claude Enterprise list controls for managing access. OpenAI says ChatGPT Enterprise supports SAML SSO, admin-set retention and a Compliance API audit log. Anthropic lists SSO, SCIM provisioning, role-based access control, audit logs, data retention controls and customer-managed encryption keys for Claude Enterprise.

Boundary 4: training and retention defaults, including feedback buttons

OpenAI says it does not train its models on business data by default, that you own your inputs and outputs where the law allows, and that it limits its staff's access to incidents, permitted recovery and legal requirements. Eligible new Enterprise customers can store content at rest in regions including the United States, the United Kingdom and Europe.
Microsoft says prompts, responses and Graph data are not used to train foundation models.
Anthropic states it does not use inputs or outputs from its commercial products to train models by default. A conversation rated with thumbs up or down can be stored for up to five years and used for training. An organization owner can turn the rating feature off. Put that toggle on your rollout checklist.
Proton says Lumo does not log chats, uses zero-access encryption so "no one (not even Proton)" can read them, and never trains on your data. Tinfoil encrypts data to a secure enclave Tinfoil itself cannot access, verifiable through hardware attestation. Venice keeps history on the device and relays requests without storing them.
Eton describes EtonAI as a private, single-tenant instance and lists ISO 42001, ISO 27001, ISO 27701, SOC 1 Type 2 and SOC 2 Type 2. Aleta lists SOC 2 Type II and encryption at rest and in transit.

The private chat apps make strong promises for a single person's conversations. For an office they are do it yourself: none of them connects to your custodians, entity records or shared drive on its own.

Concierge services

Knightsbridge Circle describes itself as "a dedicated extension of your personal office," with a personal manager who looks after four members. Fortune reported on 2026-10-11 that Aurora is an invite-only app pairing AI with human experts for members' dining, travel, luxury goods and wellness, and that members can connect calendars and wearables such as Whoop and Oura so Aurora can anticipate requests. Both fit lifestyle and travel. Neither is described as working on a family office's records, entities or investments.

The checklist

Which information should the AI reach: the portfolio, the principal's email, entity and trust documents, or all of it?
Who in your office will set it up, review permissions and keep it current?
Does the provider train on your data, and what do its feedback buttons do?
Where is the data stored, and who at the vendor can see it?
Do the answers carry citations your team can check?

Answer the first two for your own office. For the training question, policies differ by platform, so ask each vendor for its written policy.

Where Private SuperIntelligence fits

Private SuperIntelligence is an AI concierge for family offices. It works across the family's own email, documents and data, answers with citations, and acts within permissions the family sets. Each deployment is built for one client and works alongside the platforms above.

Private SuperIntelligence takes a limited number of families at a time so every one gets the same standard of service. Book a private call (ref BAF1-DEV) to talk through your office with Alex Morris.

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

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

Originally published at privatesuperintelligence.si. Vendor pages accessed 2026-10-15.

Sources

Enterprise privacy at OpenAI (OpenAI, updated 2026-01-08, accessed 2026-10-15)
ChatGPT Enterprise (OpenAI, accessed 2026-10-15)
Data residency and inference residency for ChatGPT (OpenAI Help Center, accessed 2026-10-15)
Data, Privacy, and Security for Microsoft Copilot (Microsoft Learn, accessed 2026-10-15)
Claude Enterprise (Anthropic, accessed 2026-10-15)
Is my data used for model training? (Anthropic Privacy Center, accessed 2026-10-15)
Putting AI inside investment management workflows (Addepar, 2026-03-26, accessed 2026-10-15)
AI for family offices (Masttro, accessed 2026-10-15)
Asora AI Connector (Asora, accessed 2026-10-15)
EtonAI (Eton Solutions, accessed 2026-10-15)
Aleta (Aleta, accessed 2026-10-15)
About Lumo (Proton, accessed 2026-10-15)
Tinfoil (Tinfoil, accessed 2026-10-15)
Privacy in Venice (Venice, accessed 2026-10-15)
Knightsbridge Circle (Knightsbridge Circle, accessed 2026-10-15)
Meet Aurora, the invite-only AI concierge (Fortune, Sydney Lake, 2026-10-11, accessed 2026-10-15), linked from the original article.