Ownership and Consent
CORPUS is being built in the open. Some of what you read here is live, some is still design intent — expect it to evolve.
Contributors own their works at every point in the relationship with CORPUS. The licence is a use grant, not a transfer, and it is granted explicitly per contribution — not implied by uploading, registering, or any other ambient act. This is the architectural inversion the protocol depends on: in CORPUS, training rights flow from named contributors to named uses, not from scraping or buy-out.
Opt-in, not opt-out
Every work in CORPUS enters with an explicit, AI-specific consent declaration. There is no scraping path, no implied licence, no opt-out checkbox that a rights holder must hunt down. The mechanics — collaboration agreement, named rights holders, percentage splits, vocal-consent capture — are described from the contributor's perspective in The Contribution Process.
Architecturally, opt-in matters because it sidesteps the entire DSM Article 4 reservation problem at the source: no track in the corpus relies on an unreserved TDM exception. See DSM Directive and the TDM opt-out for what that means for licensees.
Split-contribution attribution
CORPUS supports split attribution for works with multiple rights holders — composer, performer, producer, and so on. Each contributor is acknowledged, each receives their share of royalty points proportional to the agreed split, and commercial licensing requires explicit confirmation or indemnification from every named party. This is how disputes between co-rights-holders are kept transparent rather than absorbed silently into a platform's terms of service.
Withdrawal stops future use, not past compensation
Contributors can withdraw a work, or their entire body of contributions, at any time. The work is marked withdrawn in the registry and excluded from any future training runs.
What withdrawal does not do is "untrain" a model. If a work has already shaped a model, that shaping is technically embedded in the weights — there is no procedure to undo it. CORPUS handles this honestly: the work continues to be royalty-eligible for models it has already trained, and already-issued CRPS remain. The alternative — withdrawing both future use and past compensation — would punish contributors for changing their mind.
This is one of the clearest places CORPUS departs from industry norms, where ingestion typically ends both control and compensation. The contributor-side view is in Withdrawal and Rights.
Today's terms are provisional
During the public beta, contributions are governed by provisional, non-commercial terms. They permit internal model training for R&D, prototypes, and partner demos — but not the public release of trained models. At no point does contributor data become platform-owned.
Once commercial licensing opens, contributors will explicitly opt in to the commercial terms before any of their works enter revenue-generating training. Every future commercial use is opt-in and revenue-sharing; the beta consent does not silently roll over.