Source: Pinsent Masons (Out-Law)

Technology that enables online content to be identified even as it is shared, reformatted or renamed could help creators effectively opt their works out from being used for AI training, according to a European Commission‑commissioned feasibility study. The study combined desk research, stakeholder interviews, a survey and a multistakeholder workshop and concluded that existing text and data mining (TDM) opt-out mechanisms are frequently fragmented, unevenly implemented and often insufficiently effective in practice, producing heterogeneous and sometimes contradictory signals for AI developers operating at web scale.

The report found that an EU‑level registry leveraging content‑based digital fingerprinting — citing the International Standard Content Code (ISCC) — could offer a stable content‑derived handle that survives filename changes, format conversions and near‑identical copies, enabling work‑level identification across platforms. It also cautioned the registry would not act as a repository of digital assets, a licensing marketplace or an ownership adjudicator and would sit alongside existing sectoral identifiers and workflows. Pinsent Masons’ Gill Dennis said the registry could make TDM opt‑outs practically useful by addressing technical barriers to large‑scale recognition, while Cerys Wyn Davies noted the idea is not new and referred to a UK December 2024 consultation that considered a similar ‘do not train’ registry but which was later shelved amid concerns about creator burden and lack of standardisation.