What Is a Non-Exclusive Training Data License?

Non-exclusive is the term that decides whether a competitor can license the same data. It does not weaken what the buyer may do with the data, because scope of use, term and revocability are separate terms. Exclusivity matters most when a dataset’s value depends on rivals not holding it, such as a narrow product skill rather than a broad pretraining mix.

Firsthand states a perpetual, irrevocable, worldwide grant that is non-exclusive by default, covering training, evaluating, fine-tuning and deploying models, including commercial ones. Exclusivity is available per skill line or per environment, priced as a multiplier; the multiplier is not stated. Your signed contract is the document that counts, not this page.

A non-exclusive license is permission that does not stop the licensor from granting the same permission to someone else. The buyer can use the data for whatever the license allows, but nothing promises the data stays unique to them. The word describes who else may receive the data. It does not describe what the buyer may do, how long the grant lasts or whether it can be withdrawn.

Public licenses show the idea at its widest. The Creative Commons Attribution 4.0 legal code grants "a worldwide, royalty-free, non-sublicensable, non-exclusive, irrevocable license", and every recipient who accepts the terms gets the same grant. This page is general information, not legal advice.

An exclusive license is a promise that the licensor will not grant the same rights to anyone else, within a scope the contract names: a specific recording, a skill, an environment. A non-exclusive license makes no such promise. US copyright law treats them differently: the Copyright Act defines a "transfer of copyright ownership" to include an "exclusive license" and to exclude "a nonexclusive license".

Exclusivity is therefore normally its own term with a defined scope. The first table compares the two without prices. Ask counsel how your jurisdiction treats each.

The /provenance page states a perpetual, irrevocable, worldwide grant that is non-exclusive by default. Permitted use is training, evaluating, fine-tuning and deploying models, including commercial models. The term has no expiry, and a lapsed program does not revoke data already delivered.

Firsthand also states that it warrants the consent chain for every delivered episode and indemnifies against claims arising from it. The second table lists each stated term and marks what the site does not say.

The site states that derived weights are yours, with no royalty, no revenue share and no reporting obligation. It states that the footage itself cannot be redistributed, while models derived from it can be redistributed without limit.

The site describes the license as buyer-owned. It does not say whether that includes a copyright assignment, so treat that as Not stated and settle it in the signed contract.

Labs sourcing frontier AI training data should ask six things in writing. Is the grant exclusive, and over what: the footage, the skill spec or an environment? How long does any exclusivity last, and what does it cost? Is the grant perpetual and irrevocable? Does it cover trained models and derivatives? Can the data be redistributed? Is the consent behind every item documented?

Put the license scope and any exclusivity requirement in the skill spec before capture begins, then request a custom dataset. Where coverage is deep, the median time from signed spec to first validated batch is 48 hours; thinner environments run three to four weeks. The site states those times for collection, not for any licensing step.

What is a non-exclusive training data license?

A grant to use a dataset for training that does not stop the licensor from licensing the same data to others. Scope, term and revocability are separate terms.

Is non-exclusive worse for the buyer?

Not for what the buyer may do. Non-exclusive only means others can receive the same data. Firsthand states the grant is perpetual and irrevocable.

Can a license be irrevocable and non-exclusive?

Yes. The CC BY 4.0 legal code grants a license that is both non-exclusive and irrevocable.

Does Firsthand offer exclusivity?

Yes, as an added term per skill line or per environment, priced as a multiplier. The multiplier and the exclusivity duration are not stated.

Who owns models trained on Firsthand data?

The site states derived weights are yours, with no royalty, revenue share or reporting obligation.

How fast can a custom collection start?

Where coverage is deep, the median time from signed spec to first validated batch is 48 hours; thinner environments run three to four weeks.