---
title: What will investors' lawyers ask about my AI startup?
description: The legal questions due diligence asks of an AI company, including code ownership, training data, model provider terms, privacy, and customer promises, and how to have the answers ready before a raise.
category: Company
service: ai-feature-review
updated: 2026-11-17
---

## Short answer

Investors' lawyers will want to know that you own what you've built, that you had the right to use the data behind it, that your model providers' terms allow your business, and that your product does what your privacy policy and contracts say. For AI companies, the questions about data and models now get as much attention as the usual corporate housekeeping. Most of the answers already exist in your codebase and contracts. Gathering them before the raise makes diligence faster and keeps surprises from turning into lower valuations or special conditions in the deal.

## What they'll ask about

### Who owns the code

- **IP assignments.** Every founder, employee, and contractor who wrote code should have signed an agreement assigning it to the company. Missing assignments from early contractors or co-founders who left are among the most common diligence problems.
- **AI-written code.** US copyright requires a human author, and the Supreme Court declined in March 2026 to revisit that rule. Code generated entirely by an AI tool may not be protected by copyright, though the human work around it, such as design, selection, and editing, can be. See [Who owns AI-generated output?](/guides/who-owns-ai-outputs)
- **Open source.** Dependencies, and code an AI tool suggested, can carry license obligations. Copyleft licenses can require sharing your source code if you distribute the software. Expect a license scan.

### Training data and models

- **Where the data came from.** If you trained or fine-tuned models, diligence will ask for the sources and your right to use each one. The stakes became concrete in 2025 and 2026. In *Bartz v. Anthropic*, the court found that training on lawfully acquired books could be fair use but that keeping a library of pirated books was not, and a $1.5 billion class settlement won final approval in July 2026. In September 2026, the Third Circuit held in *Thomson Reuters v. Ross* that copying a competitor's legal headnotes to train a non-generative AI tool wasn't fair use. Courts are still working out how these rules apply to generative AI.
- **User data.** If customer data trained a model, diligence will ask whether your terms and privacy policy allowed it. See [Can I train or fine-tune a model on my users' data?](/guides/training-models-on-customer-data)
- **Model provider terms.** Expect questions about which AI providers you use, what their terms allow, what they do with your data, any limits on building competing models, and what happens if a provider changes its terms or prices.

### Privacy and AI law compliance

- **The privacy policy matches the product.** Diligence compares what your privacy policy says with what your product does, including data sent to AI providers, analytics and session replay, and retention and deletion.
- **AI-specific laws.** State AI laws are arriving quickly, including Colorado's automated decision law, California's automated decision rules, and companion-chatbot laws. Expect questions about which apply to you.
- **Sensitive data.** Health data, financial data, children's data, and biometrics draw extra scrutiny.
- **Incidents.** Security incidents, breach notices, and regulator inquiries will be asked about directly.

### Customers and claims

- **What you promised.** Enterprise contracts often include promises about not training on customer data, where data is stored, uptime, and indemnities. Diligence checks whether the product keeps them, and whether any liability is uncapped.
- **Marketing claims.** Claims about accuracy and capability can create FTC exposure. See [What can I claim about my AI product in marketing?](/guides/marketing-claims-for-ai-products)
- **Platform terms.** App store rules and the terms of integrations like Google Workspace and Slack can restrict what your product does with data. See [Can my AI app connect to users' Gmail, Google Drive, or Slack?](/guides/connecting-ai-apps-to-gmail-drive-slack)

## How to get ready

- **Collect signed IP assignments** from everyone who has contributed code, and fix any gaps before the raise.
- **Run a license scan** of dependencies and keep the report.
- **Write down your data provenance**: each training and fine-tuning dataset, where it came from, and your right to use it.
- **List your AI providers** and the plan, settings, and terms you're on, including data retention and training settings.
- **Check your privacy policy against the code**, and fix what doesn't match.
- **List customer contracts** with non-standard promises, especially about data, training, and indemnities.
- **Keep an incident log**: security incidents, complaints, and regulator contacts, with how each was handled.

A clean data room shortens diligence and puts you in a stronger position when terms are negotiated.

## Checklist an agent can run against the codebase

- [ ] List every third-party dependency with its license, and flag copyleft licenses.
- [ ] List every AI model provider and API the product calls, with the account settings for training and retention where configured.
- [ ] Find any training, fine-tuning, or evaluation pipelines, and list their data sources.
- [ ] Find where customer or user data could reach a training or evaluation set.
- [ ] List every place user data is stored or sent, for comparison with the privacy policy.
- [ ] Find secrets or credentials committed to the repository's history.
- [ ] List contributors in the repository's history, for comparison with signed IP assignments.

## Sources

- [U.S. Copyright Office: Copyright and Artificial Intelligence](https://www.copyright.gov/ai/)

## If this applies to you

Before a raise, an **AI feature review** checks the AI-specific items diligence will focus on: model provider terms, data flows, training data, privacy documents, and AI-law compliance, with documents you can put in the data room. For a single question, such as whether a planned training dataset is safe to use, a **pre-build feature check** covers that decision before you build.
