Guide
AI data licence terms: exclusivity, scope and what to negotiate
Short answer: The key AI data licence terms are exclusivity (whether you can license the same records to anyone else, and for how long), scope of use (training only, or more), who may use it (the buyer only, or sub-licensees), duration, de-identification, security and deletion, and payment timing. Exclusivity is increasingly what buyers ask for: time-limit it, narrow it to AI training, and keep your right to use records for non-AI purposes.
The clauses that decide what you're really selling, in plain English. General information, not legal advice.
Last updated · Sources linked inline · Buyer figures as published, observed Oct 16, 2026 (micro1, Mercor and GDC re-checked Nov 3, 2026)
What does exclusivity mean in an AI data licence?
You agree not to license the same records (or similar ones) to anyone else for a period, a field or a territory.
A non-exclusive licence lets you license the same copy to several buyers. An exclusive licence gives one buyer sole rights, usually for a period and a defined use. Published programs differ: Labelbox says its licences are exclusive by default; SuperAnnotate says it pays on delivery and on each relicense; Scale says partnerships can recur. As published; verify each agreement.
Why do AI buyers ask for exclusivity?
Because data that every lab has gives no lab an edge.
A broker commentator puts it bluntly: "exclusivity is king right now", adding that "some labs won't even entertain a dataset if it's already being shopped around" (post; opinion). That's also why we introduce a firm to one fitting buyer at a time rather than circulating its details.
What should a firm negotiate?
A time limit, a narrow field of use, a defined set of users, and your right to keep using your own records.
Polyshares' co-founder suggests "at most 2 years of exclusivity", "very limited scope of use for the purpose of AI model training", and that you "protect your ability to monetize your data to non-AI buyers" (post; opinion; his post also limits use to US-based labs, which may or may not suit you). Another market recap describes perpetual use as common but exclusivity as time-limited, around 2 years (post; opinion). Treat these as starting points, not market standards.
Which clauses should you check before signing?
| Clause | Question to ask |
|---|---|
| Exclusivity | Exclusive or not? For how long, for what use, and over which records? |
| Permitted use | Training only, or evaluation, fine-tuning, products, resale? |
| Sub-licensing | Can the buyer pass the data to other labs or customers? Who are they? |
| Duration | Is the licence perpetual or for a term? What happens at the end? |
| De-identification | Who does it, to what standard, and who carries the risk if it fails? |
| Security and deletion | Where is data stored, who can access it, and when is it deleted? |
| Payment | Fixed or milestone? When is it due (e.g. on delivery, after review)? Any relicense share? |
| Your own use | Can you still use your records internally and with non-AI buyers? |
| Intermediary fees | Is any agent or intermediary fee disclosed in writing? |
Why not send records to every buyer at once?
Because a dataset that's widely shopped can lose value, and widely shared samples raise confidentiality risk.
Share an inventory first, samples only under NDA, and one buyer at a time where you can. The intake form asks whether your records have been shown or licensed elsewhere so a buyer gets an honest picture up front.
Sources: termsheetinator (X) · Polyshares co-founder on terms (X) · indexsy recap (X) · buyer terms via the Who Sells Data register (observed Oct 16, 2026)
Related: how to prepare your records · privacy and compliance
Part of our guide series. Start with the pillar: how to license your firm's records to AI.
What do firms ask about licence terms?
Is an exclusive AI data licence better?
How long should exclusivity last?
Should I sign a perpetual licence?
Can I license the same records to two buyers?
What could your firm's records be worth?
Three answers give you a rough range based on what buyers publish. The rest helps a buyer take you seriously: every question after step 1 is optional. No files, logins or client names.