About the GPAI Ledger

The GPAI Ledger is a public-interest archive of the training-data disclosures the EU AI Act requires: Article 53(1)(d) obliges providers of general-purpose AI models on the EU market to publish a summary of the content used to train each model. Providers publish these documents on their own websites and may change, move, or remove them at any time; no official registry exists. This ledger checks every known summary on a daily schedule, keeps every content version, and records the evidence needed to prove what was published when. Every check is logged in the append-only event log, so any missed sweep is itself visible in the record.

Who runs it

The GPAI Ledger is an independent, self-funded project. It is not affiliated with any AI provider, law firm, or regulator, and accepts no funding from the companies it tracks. Model metadata is seeded from the AI Accountability Lab (AIAL) at Trinity College Dublin, with attribution, plus this project's own monitoring and discoveries.

Contact

Corrections, disputes, provider objections, or tips about summaries we missed: contact@gpailedger.com. Providers who object to full-text archiving of a document are offered a structured-facts treatment — hashes, sizes and provenance metadata without the document bytes; see the methodology page.

What is automated, what is human

Capture, hashing, timestamping, and page generation are fully automated and run on a public, auditable pipeline. Which sources to track, how to label their status, and every editorial note are human decisions. Errors are corrected openly — see the corrections log.

Relationship to AIAL

AIAL grades the quality of published summaries and maintains the tracker whose metadata seeds this ledger's model list. AIAL's grades and scoping assessments are theirs; this ledger's captures, version chains, and provenance are this project's. Their pages are archived here so the provenance chain includes its own upstream source.

Licensing

Capture manifests, hashes, and the append-only event log are released under CC0. Archived documents remain the property of their publishers and are served for public-interest transparency, research, and verification, with attribution.