Most founders shipping AI features have never asked the one question their first enterprise customer will: who owns what comes out of the model? It feels academic right up until it’s a redline in a deal you need to close.
The good news is that the answer is usually simple once you’ve decided it on purpose. The bad news is that “on purpose” is the part founders skip. Here’s the short version of what to nail down before you ship.
1. Output ownership
When your product generates something — text, an image, code, a summary — your terms should say plainly who owns it. For most B2B products the cleanest answer is: the customer owns their output, and grants you a license to operate and improve the service. Vague silence here is what turns into a negotiation later.
- State who owns generated output, and any license each side grants.
- Carve out aggregated, de-identified data if you rely on it.
- Be explicit that customers are responsible for how they use output.
Silence in your terms is a decision — usually the wrong one.
2. Training data & customer content
Your customers want to know one thing above all: are you training on their data? If the answer is no, say so clearly — it’s a feature. If the answer is “only in aggregate” or “only with opt-in,” define those terms precisely. Enterprise buyers will hold you to the exact words.
This is also where your relationship with your model provider matters. If you call a third-party model, your terms can’t promise more than your provider’s terms allow. Read up the stack before you write down the promise.
The provider chain
Map it once: your customer → you → your model provider. Confidentiality, data use and liability all have to flow consistently along that chain, or you’ve signed up for a gap you’ll personally own.
3. IP indemnity
Increasingly, enterprise customers ask AI vendors to indemnify them if a model’s output infringes someone’s IP. Whether you can offer that — and how you cap it — depends entirely on your provider stack and your insurance. Decide your position before it’s a term-sheet surprise, not after.
The founder’s move
You don’t need a 40-page AI policy to ship. You need four sentences you’ve actually decided on: who owns output, whether you train on customer data, how confidentiality flows to your provider, and what you’ll stand behind on IP. Get those right and most customer redlines evaporate.
When you’re ready to put it in writing, have it reviewed — AI to flag the gaps, an attorney to sign off. That’s the whole point: move fast, but don’t ship the part that bites.