What US Publishers Need to Do Before August 2, 2026
Most publishers know their content has been used to train AI models. Fewer know they had a legal right to prevent it — and that the enforcement deadline is now five months away.
Publishers have held the legal right to opt out of AI training for several years. The EU’s Copyright in the Digital Single Market Directive — in force since 2019 — gave rights holders a clear mechanism: reserve your works in machine-readable form, and AI developers are legally required to exclude them from training datasets.
The problem is that this right has been largely theoretical. There is no standardized infrastructure for declaring opt-outs that AI developers are required to check. A robots.txt entry, a metadata tag, a policy page — these are gestures toward compliance instead of enforceable records.
On August 2, 2026, the stakes change.
That is the date the EU AI Act’s enforcement powers come into full force for General Purpose AI model providers. Fines reach up to 3% of global annual turnover.
OpenAI, Google, Meta, Mistral, and every other developer placing a GPAI model on the EU market becomes legally obligated to detect and honor declared copyright reservations. The compliance obligation applies to any model offered on the EU market, regardless of where training data was sourced.
US publishers are not exempt
If your titles were ingested into a model that serves European readers — and the probability is significant — your rights declarations need to be in place before the deadline. The regulation does not change no matter where you are headquartered.
Collective agreements cannot solve edition-level rights
Collective opt-in agreements — the kind being developed through publishers’ associations and licensing bodies — address title-level rights. They do not answer a more specific question: which exact version, edition, or file of a given title was ingested by which model, and when?
This matters because enforcement is going to require evidence, not intent. A declaration made in 2026 cannot retroactively protect content ingested in 2024. A timestamped, content-specific record, keyed to the digital fingerprint of the publication's file, is what makes a rights declaration auditable.
ISCC (ISO 24138:2024) is the technical standard designed for this. It generates a content-derived fingerprint for any piece of content. This digital marker persists even when a file is reformatted or repackaged, making it possible to verify whether a specific version of a specific title is present in a training dataset.
What rights and digital strategy leads should be doing now
The window to establish a defensible record ahead of the August 2026 deadline is the next few months. The right questions to be asking now: What do current rights declarations look like in machine-readable form? Are they consistent across formats and platforms? Do they identify specific editions rather than just titles?
Amlet’s Global AI Content Registry is the registration layer designed to answer those questions. Amlet offers a neutral, standards-based infrastructure that lets publishers declare once and create a timestamped, auditable record of that declaration. Early registration is free at the base tier, and it establishes the kind of pre-deadline record that may matter considerably if enforcement becomes contested.
The era of scraping without consequence is ending. The question now is whether your rights declarations are in a form the new enforcement regime can recognize.