Since 2 August 2026, Article 50.2 of Regulation (EU) 2024/1689 requires machine-readable marking of text that an AI system generates or manipulates.
Collect produces catalog metadata with AI, and that output falls under the rule: Collect marks that output. Whoever runs the system can show the mark to an auditor or in a procurement file, without changing the MARC21, ISAD-G or the other formats they already use.
That fits what we already said in why trust a catalog made with AI: the model completes the record; trust sits in being able to say where each piece of data comes from.
What the institution gets
- A machine-readable mark on Collect’s new output, to the extent Article 50.2 requires.
- The catalog record does not change: the contract with the ILS, exporters and the public catalog stays the same.
- The proof stays with the institution and does not depend on the cataloguing process being up at that moment.
- If the institution already signs documents (NOM-151 / eIDAS), it can seal that mark with the same certificate. The signature is not a requirement of the article; it is an option.
A statistical watermark, built for running prose, is the wrong instrument for short, structured catalog fields. Collect’s mark is built for that kind of output.
Limits
Article 50.2 does not replace human review. If the institution has editorial control over the record, that is its own flow; Collect does not pretend to detect it.
The mark applies to new output: we do not rewrite the already processed holdings.
Implementation details — schema, storage, classification rules — belong to the product and are documented for whoever already runs Collect, not on this blog.
If your institution needs evidence of this compliance for an auditor or a tender, write to us at info@janium.com.