AI Usage License v1
When you sign a file with Reconize and turn on Reserve AI rights (no training), Reconize records a standards-based AI-usage preference inside the file's Content Credentials. This page is the human-readable reference that the machine-readable preference points to. It explains what that preference asks for — and, importantly, what it does and does not do.
A request, not a lock. This preference is a declarative, machine-readable request that AI systems honor voluntarily. Reconize cannot technically prevent anyone from training on or copying a file. It records your stated preference in an open, standards-based form so that operators who choose to respect creators' wishes — and, increasingly, regulators — can read and act on it. It is not a technical restriction, a DRM lock, or a guarantee.
1. What is recorded
Reconize writes the CAWG Training and Data Mining assertion (cawg.training-mining), a standard defined by the Decentralized Identity Foundation's Creator Assertions Working Group and carried inside C2PA Content Credentials. For a file signed with this preference on, Reconize sets:
- Generative AI training (
cawg.ai_generative_training): not allowed. - AI/ML training (
cawg.ai_training): not allowed. - Data mining (
cawg.data_mining): not allowed. - AI inference (
cawg.ai_inference): constrained, under the conditions in Section 2 below.
"Not allowed" means the creator does not grant permission to use the file for that purpose. It is a statement of the rights the creator reserves, expressed in a form machines can read at scale.
2. What "constrained inference" means
Inference is set to constrained rather than fully disallowed because some transient, non-training uses of a file are ordinary and useful, while others are not. Under this license, running an AI model over a file signed with this preference is permitted only when the use is:
- Transient and non-retained — the file, and any features or embeddings derived from it, are used in the moment and not stored to build or improve a model or a dataset;
- Non-training — the file is not used, directly or indirectly, to train, fine-tune, distill, align, or otherwise adjust the weights of any model;
- Provenance-, safety-, or accessibility-oriented — for example: reading and verifying Content Credentials, detecting abuse or CSAM, generating alt-text or captions for accessibility, or indexing for search that links back to the original; and
- Attribution-preserving — the operation does not strip or obscure the file's Content Credentials, and any surfaced result preserves attribution to the creator.
Inference that produces a substantially similar derivative work, that builds a reusable dataset or index of the content itself, or that contributes to model training in any form is outside this constraint and is not permitted.
3. Precedence and other terms
This license expresses a default preference recorded at signing time. Where a creator has a separate, written agreement with a specific party, that agreement governs between them. Nothing here grants any license to the underlying content beyond the limited inference use described in Section 2; all other rights are reserved by the creator. This page does not transfer copyright, and it does not create a contract between Reconize and any reader.
4. Honesty about enforcement
Reconize is a provenance tool, not an enforcement mechanism. The preference recorded here is only as effective as the willingness of AI operators to honor it and the legal frameworks that give it weight. Reconize makes no representation that any particular operator will comply, and turning this preference on does not by itself prevent training, copying, or misuse. It does give you a clear, timestamped, standards-based record of the rights you reserved at the moment you signed — which is what makes the request legible and, where the law requires operators to respect such signals, enforceable by others.
5. Changes
This is version 1 of the AI Usage License. If the terms change materially, we will publish a new version at a new URL so that files already signed continue to reference the exact terms in effect when they were signed. Files that reference this page always refer to Version 1.
6. Contact
Questions about this license: hello@reconize.me. Background on the standard is at the Creator Assertions Working Group and in our AI disclosure and EU AI Act notes.