ChatGPT at a venture firm: an engineer's checklist for deal data
Disclosure: I co-founded Mitosis Labs, which runs Private SuperIntelligence. This is an adaptation of the original guide for people who build or run internal tools at venture firms. General information, not legal advice.
If you're the person a venture firm asks to "set up AI," you inherit three constraints nobody wrote down: the provider's data terms, the founders' NDAs and the LPs' due diligence questionnaires. Here's how they map to config decisions.
1. Training defaults are per plan, not per vendor
From OpenAI's own documentation:
- Personal Free, Plus and Pro: data sharing for training is on by default. Users can turn off "Improve the model for everyone." Thumbs up or down feedback can still send the whole conversation for training.
- Business, Enterprise, Edu: OpenAI says it does not train on workspace content by default.
- API Platform: not used for training by default. Abuse monitoring logs are kept up to 30 days by default. Zero Data Retention requires OpenAI's prior approval.
- Enterprise: admins set retention; deleted conversations leave OpenAI's systems within 30 days unless law requires otherwise.
The config takeaway: the risk is often shadow usage on personal plans, not the vendor you procured. Inventory accounts before you tune settings.
2. Retention is a policy, and policies can be suspended
In 2025 OpenAI was ordered to preserve consumer ChatGPT and API content it would otherwise delete, in the New York Times copyright case. OpenAI listed Free, Plus, Pro and Team (now Business) subscriptions as affected. It says the obligation ended September 26, 2025, and that Enterprise, Edu and ZDR API traffic was excluded.
In January 2026 a federal judge affirmed orders to produce a sample of 20 million de-identified chat logs.
Design as if a third party's lawsuit can pause your vendor's deletion schedule. Enterprise, Edu and ZDR traffic sat outside that order; not storing the data at the provider at all is the answer that does not depend on the next one.
3. The NDA is an access-control policy
A June 2026 DarrowEverett analysis argues NDAs that restrict disclosure to "third parties" may be breached by sending deal information to an AI provider, "even if an NDA is silent on AI."
It lists four clause types appearing in new NDAs: express prohibitions, expanded definitions of disclosure that include "automated systems, machine learning models, or AI-powered platforms," permitted-use carve-outs for isolated no-training tools, and representations about internal AI policy.
Treat each as a rule your tooling enforces:
data_class: founder_dataroom_strict_nda-> no external modeldata_class: founder_deck_standard_nda-> reviewed business workspace onlydata_class: lp_records / side_letters-> blocked until counsel clearsdata_class: public / own_notes-> any approved tool
4. The DDQ is your acceptance test
ILPA's DDQ 2.0 doesn't name AI, but 18.14 asks whether data held by third-party service providers gets security reviews and independent assessments, 18.15 asks about encryption at rest and in transit, and Appendix D templates the third parties and technology tools you use.
ILPA's Responsible AI Quick Guide (January 2024) adds: "What are the GP's controls around AI use at the fund level?"
If your setup can't produce a one-paragraph answer naming the tool, the plan, the training default, provider access and your data rule, it isn't done.
Where we fit
Private SuperIntelligence is run by Mitosis Labs. Each deployment is built for one firm, works across its own email, documents and data, and answers with citations; 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 (ref CVC1-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: OpenAI, Data controls in ChatGPT, https://help.openai.com/en/articles/7730893-data-controls-faq ; OpenAI, Enterprise privacy at OpenAI (updated January 8, 2026), https://openai.com/enterprise-privacy/ ; OpenAI, ChatGPT Business General FAQ, https://help.openai.com/en/articles/8542115-chatgpt-business-faq ; OpenAI, How we're responding to The New York Times' data demands (updated October 22, 2025), https://openai.com/index/response-to-nyt-data-demands/ ; ABA Journal, ChatGPT creator must turn over 20M chat logs (January 8, 2026), https://www.abajournal.com/news/article/chatgpt-creator-must-turn-over-20m-chat-logs-in-copyright-litigation-federal-judge-says ; DarrowEverett LLP, Loose Lips, Large Language Models: What Your NDA is Missing in the Age of AI (June 2, 2026), https://darroweverett.com/ndas-non-disclosure-agreements-age-ai-practical-steps-analysis/ ; ILPA, Due Diligence Questionnaire 2.0 (November 2021), https://ilpa.org/wp-content/uploads/2021/11/ILPA-DDQ-2.0.pdf ; ILPA, Responsible AI Quick Guide for Asset Owners (January 2024), https://ilpa.org/wp-content/uploads/2023/10/Responsible-AI-Quick-Guide-for-Asset-Owners.pdf
Comments
No comments yet. Start the discussion.