How AI Labs Keep Custom Training Data Exclusive

Exclusive training data means no other lab is licensed to train on the same footage or, ideally, on data captured to the same skill spec. That is a narrower and stronger claim than "custom" or "commissioned" — a vendor can build a dataset to your spec and still resell equivalent data captured under a similar spec to a competitor, unless the contract says otherwise. Exclusivity has to be a term the buyer actually holds, not an assumption made because the vendor built it for them.

This describes how exclusivity is typically structured in a custom data collection contract. It is general information, not legal advice — confirm the specific terms of any order form before relying on them.

A public dataset, by definition, is available to every lab that wants it — anything trained on it is available to competitors on the same terms. A shared or off-the-shelf commercial dataset is usually licensed non-exclusively to many buyers at once, so the underlying footage is not unique even when the license is commercial. Neither gives a lab a data advantage its competitors cannot also buy. A custom collection only creates that advantage if the resulting data and the spec that produced it are exclusive to the commissioning buyer.

A buyer-owned license is a perpetual, irrevocable commercial license granted directly to the buyer, rather than a revocable subscription the provider can resell or reassign. On its own, buyer-owned licensing settles who can train on the delivered footage and for how long — it does not by itself stop the vendor from capturing similar data under a similar spec for a different buyer. That is why exclusivity needs its own clause on top of the license terms, covering the footage, the spec, or both, depending on what the buyer wants to protect.

See /guides/buyer-owned-training-data for what a buyer-owned commercial license actually grants, and how it differs from a research-only or revocable license.

A documented consent chain — a signed release per participant, a recorded legal basis, and a per-episode artefact trail — is what makes the buyer-owned license enforceable rather than assumed. If a competitor later claims a right to use overlapping footage, the data card and consent artefacts are the record that shows exactly what was granted, to whom, and when. Provenance documentation does the same job for the dataset as a whole: it records what was captured, under what terms, and for which buyer, so exclusivity is checkable rather than taken on trust.

See /provenance for the full consent chain and licensing record, and /guides/dataset-provenance-explained for what a provenance record needs to include.

De-identification — detecting and irreversibly blurring faces, plates, and screens, and scrubbing identifying speech — protects the people captured, not the buyer’s exclusivity claim directly. But it matters for exclusivity in a related way: footage that still contains identifiable bystanders or unredacted environments is harder to license exclusively, because the underlying consent and privacy exposure follows the footage wherever it goes. A dataset that is properly de-identified and consent-documented is the version a buyer can actually hold exclusively without inheriting someone else’s privacy risk.

An exclusivity clause should say, in plain terms, who else — if anyone — can receive the same footage or data captured to the same or a substantially similar skill spec, for how long, and in which markets or use cases. It sits alongside the license grant and the consent chain rather than replacing either of them: the license says what the buyer can do with the data, consent says the people in it agreed, and exclusivity says nobody else gets the same or equivalent data.

• Does the license cover the delivered footage only, or does it also restrict the vendor from capturing similar data to the same spec for another buyer?
• Is the commercial license perpetual and irrevocable, or can it be revoked or resold? See /guides/buyer-owned-training-data.
• Is there a documented consent chain that can be produced if exclusivity is ever contested? See /provenance.
• Is exclusivity time-bound, and if so, what happens to the restriction once it expires?
• Does the exclusivity clause name specific markets, use cases, or competitors, or is it a blanket restriction?
• What rework or rejection terms apply before delivery, so the exclusive data actually meets the buyer’s acceptance criteria? See /guides/robotics-data-collection-sla.

Starting an exclusive custom collection means writing the exclusivity terms into the skill spec and order form before capture begins — not asking for them after the data has already been delivered. That means specifying the license (buyer-owned, perpetual, irrevocable), the consent and provenance documentation the buyer will receive, and the scope and duration of the exclusivity restriction itself, alongside the usual scope, task, and acceptance criteria in the spec.

See /custom-collection/frontier-ai-labs for how custom collection is structured for frontier labs, and /guides/robotics-data-collection-sla for what else belongs in the contract.