Herndon and Dryhurst built the opt-out layer AI training lacked

The system

Spawning is a technology organisation, associated with musicians Holly Herndon and Mat Dryhurst, that builds infrastructure letting artists register consent, or its absence, over the use of their published work in AI training datasets. It is separate from Holly+, the pair's vocal-deepfake performance instrument built on Herndon's own voice; Spawning is opt-out and licensing infrastructure meant to apply across many creators' catalogs, in any medium. Contemporaneous reporting from September 2022 describes Dryhurst and Herndon as 'spearheading efforts such as Source+, a standard aiming to allow people to disallow their work or likeness to be used for AI training purposes,' alongside the company's image-search tool, Have I Been Trained.

What the documents establish

Spawning's own technical documentation for its datadiligence tool states the problem directly: 'ML training datasets are often harvested without consent from the data or content owners,' and that, absent a shared opt-out standard, 'many platforms and individuals have come up with their own methods of stating their consent.' The tool's stated purpose is to give developers 'a consistent interface to check if a given work is opted-out using any known method' before training on it, and its documentation lists specific mechanisms it respects, including the Spawning API, DeviantArt's opt-out header, and C2PA and CAI content-credential metadata. The 2022 report independently confirms the same period and framing: a public search tool over image training data, paired with a developing opt-out standard, rather than a paid licensing marketplace at that stage.

Craft and rights

What this infrastructure gives a musician is a place to state a preference before training happens, rather than only a complaint afterwards. That is narrower than enforcement: the datadiligence documentation frames opt-out compliance as something a training pipeline must choose to check, not something a data format enforces on its own, and Spawning's current site still asks 'AI trainers looking to respect the Do-Not-Train registry' to reach out directly, which implies voluntary adoption rather than a legal mandate. Registering a catalog with this kind of registry is a low-cost hedge for a working artist, but it does not stop a model already trained on unlicensed data, and it is a different instrument from a licensing deal that pays for confirmed use.

Outcomes and open questions

Public repository activity under the Spawning organisation shows continued work on a Source.Plus product as recently as late 2025, suggesting the opt-out registry is evolving toward a paid, opt-in licensing marketplace, though the public-facing site was not fully reachable at the time of writing.

• Which AI trainers have publicly committed to checking Spawning's Do-Not-Train registry before training?
• Has the opt-out registry converted into a functioning paid-licensing marketplace, and on what terms?
• How does registering with Spawning interact with a platform's own separate terms of service over the same work?

Voluntary opt-out infrastructure is a meaningful step in a market with no binding standard, but it is worth reading Spawning's own materials for what they say they do, rather than the broader press shorthand of 'AI training consent.'