What Is a Participant Release Form for AI Training Data?

A participant release form is the contributor’s side of the paperwork: the person in the footage agrees in writing to be recorded and to the intended use. A data licence is a different document, the grant that defines what a buyer can do with the dataset. The release is what the licence rests on, so ask who signed what, when, and whether proof exists per episode. This page summarises what the site states and is not legal advice.

Every Firsthand contributor signs a written release granting commercial training rights before capture, with identity verified and an hourly rate agreed in Schedule A. Each batch ships with consent artefact IDs listed per episode in the data card, which also lists jurisdictions of capture. Firsthand warrants the consent chain for every delivered episode and indemnifies against claims arising from it. The release text is published verbatim on the provenance page. Jurisdiction-specific form terms and the signature method are Not stated.

The release text on the provenance page (version 3.2, effective 2026-01-14) grants a perpetual, irrevocable, worldwide, sublicensable, royalty-free right to record, store, process and distribute the audio, video, depth and motion recordings from the sessions in Schedule A, and to license them to third parties to train, evaluate and deploy models, including commercial models.

The grant extends to derivatives: depth maps, hand and body pose estimates, segmentation masks, textual action labels and model weights. It does not license a name, likeness or voice for advertising or endorsement.

The release is the first of six checkpoints: written release and identity check before capture; site permission at capture; face, plate and screen detection at ingest; human review of every frame; consent artefact IDs per episode at delivery; and, after delivery, withdrawal honoured for undelivered footage only. The site states that a break at any checkpoint discards the footage rather than downgrading it. The release proves one link; the chain traces each episode back to a signature.

A release covers the person who signs it. For bystanders, the participant confirms they have not knowingly recorded anyone who has not signed, and Firsthand de-identifies any unidentified person or discards the footage. Where a use case needs identifiable people, the site says it scopes a capture in which every subject has signed a release.

Licensing is the next layer: data ships under a perpetual, irrevocable, worldwide licence that is buyer-owned and non-exclusive by default, with derived weights belonging to the buyer.

The Ego4D consortium’s privacy and ethics statement says camera-wearers could withdraw at any point and that consent forms or video release forms were signed, with each university or organization keeping the record. That describes Ego4D only, not Firsthand. Whether it permits a given commercial use is a licence question to check directly.

Write down the tasks, environments and objects your model needs, state that every contributor must have signed a release, and name the delivery format. Most orders begin with a Spec Pilot, a small, fast first batch that proves the spec on real footage. This is on-demand custom collection, not a catalog download.

Paid contributors perform the task; Firsthand does not capture at a customer’s own facility. Where coverage is deep the median from signed spec to first validated batch is 48 hours; where it is thin, three to four weeks. Billing is per validated hour, with no list price.

What is a participant release form for AI training data?

The signed document in which a person appearing in a recording grants the rights needed to use it for AI training. At Firsthand it is signed, with identity verified, before capture.

Is a release form the same as a data licence?

No. The release is the contributor’s grant. The data licence is the separate grant that defines what a buyer can do with the delivered dataset.

Can a contributor withdraw after delivery?

They can withdraw from future sessions at any time and request deletion of recordings not yet delivered. Delivered recordings cannot be recalled, and the release says so.

Do I receive proof that releases exist?

Each batch lists consent artefact IDs per episode in the data card. Firsthand warrants the consent chain for every delivered episode.

Is this page legal advice?

No. It summarises statements published on this site. The executed agreement controls, and your counsel should review it.

How fast can a consented custom dataset be collected?

Where coverage is deep, the median from signed spec to first validated batch is 48 hours; where it is thin, three to four weeks. There is no list price.