GPAI disclosures: the Art. 53/55 workspace
For general-purpose model providers: downstream model documentation, the copyright policy, the training-data summary — and the extra Art. 55 layer when a model carries systemic risk.
Documentation → GPAI Disclosures collects the duties that attach to general-purpose AI models — Articles 53 and 55 — across your portfolio. For what counts as GPAI in the first place, and whether these duties are yours at all, start with What counts as GPAI, and what you must do in the EU AI Act chapter; this article is about doing the work in Veritome.
Which systems appear
Systems classified with a GPAI role during classification. Each row rolls up that system's GPAI obligations; click through to the system's GPAI tab to work them.
The Art. 53 set
- Model documentation for downstream providers — what an integrator needs to build on your model responsibly: capabilities, limitations, intended tasks, evaluation results. The engine's checklist mirrors the Annex XI/XII structure.
- Copyright policy — a stated policy honouring TDM opt-outs, kept as a living document (it makes a natural evidence record — see The Evidence hub).
- Training-data summary — the public summary in the AI Office template.
Open-source models get partial relief from Art. 53, but not from the copyright policy or the training-data summary — the checklist reflects that split rather than waving the whole article.
The Art. 55 layer — systemic risk
If a model crosses the systemic-risk threshold (the compute presumption, or designation), a second layer of obligations appears automatically: model evaluations including adversarial testing, systemic-risk assessment and mitigation, serious-incident tracking, and cybersecurity protection. The engine adds these only when the classification says so — you don't opt in or out by hand.
Timing
GPAI duties have applied to new models since 2 August 2025, with models already on the market before that date given until 2 August 2027. The Calendar places both against your systems.