Why this object hangs off that object

Maintain technical documentation, information for downstream providers, a copyright policy and a public summary of training content.

Hangs off: Article 53: GPAI model providers

Editorially reviewed | gpai

Build one package for providers integrating your model, covering the intended tasks and integration options, acceptable use policies, release date and distribution methods, interaction with external hardware or software, software versions, architecture and parameter count, modality and format of inputs and outputs including maximum size, licence, required technical means, and information on the training, testing and validation data used.

Hangs off: Article 53: GPAI model providers

Editorially reviewed | gpai

Record which techniques you use to identify a reservation of rights within the meaning of Article 4(3) of Directive (EU) 2019/790 when collecting training data, how often you recheck, and how you then comply with that reservation.

Hangs off: Article 53: GPAI model providers

Editorially reviewed | gpai

The voluntary code of practice gives GPAI model providers a route to demonstrate compliance.

Hangs off: Article 53: GPAI model providers

Placed against the official source | gpai

The Commission explains when someone becomes the provider of a GPAI model, including through fine-tuning.

Hangs off: Article 53: GPAI model providers, Article 55: GPAI models with systemic risk

Placed against the official source | gpai

Since 2 August 2025 the obligations for providers of general-purpose AI models apply.

Hangs off: Article 53: GPAI model providers, Article 55: GPAI models with systemic risk

Placed against the official source | gpai, timeline

Since 2 August 2026 the Commission can request GPAI information, conduct evaluations and require measures.

Hangs off: Article 53: GPAI model providers

Placed against the official source | enforcement, gpai

Models placed on the market before 2 August 2025 have until 2 August 2027.

Hangs off: Article 53: GPAI model providers, Article 55: GPAI models with systemic risk

Placed against the official source | gpai, timeline

Establish every six months whether you demonstrate compliance through a code of practice, through a published harmonised standard, or through alternative adequate means, and whether the underlying file matches that choice.

Hangs off: Article 53: GPAI model providers

Editorially reviewed | gpai

Update documentation and downstream information when the model, capabilities or risks change.

Hangs off: Article 53: GPAI model providers

Editorially reviewed | control, gpai

Current technical documentation, downstream information, copyright policy and public training summary.

Hangs off: Article 53: GPAI model providers

Editorially reviewed | evidence, gpai

A register with one row per model version: release date, distribution method, architecture and parameter count, compute used, and a note of which Annex XI and Annex XII elements changed relative to the previous version.

Hangs off: Article 53: GPAI model providers

Editorially reviewed | gpai

The published, sufficiently detailed summary of the content used to train the model, drawn up according to the template provided by the AI Office, with publication date and the corresponding model version.

Hangs off: Article 53: GPAI model providers

Editorially reviewed | gpai

A European scale-up fine-tunes an existing general-purpose AI model for its own product, using more compute than one third of the compute used to train the original model.

Hangs off: Article 53: GPAI model providers

Placed against the official source | examples

A bank fine-tunes an existing general-purpose AI model on its own product documentation and customer questions, using a fraction of the original compute, and builds a customer chatbot around it that it offers under its own name.

Hangs off: Article 53: GPAI model providers

Placed against the official source | examples

A downstream actor modifies an existing systemic-risk model so substantially that the change exceeds the threshold, and publishes the result as its own model.

Hangs off: Article 53: GPAI model providers

Placed against the official source | examples

Three providers call their model open source. The first licence allows non-commercial research only. The second requires a separate commercial licence once monthly active users pass a threshold. The third gives the model away for free but hosts it exclusively on its own platform where visitors are served paid advertisements.

Hangs off: Article 53: GPAI model providers

Placed against the official source | examples

A research group publishes a general-purpose AI model under a free and open-source licence and makes the weights, architecture and usage information publicly available. The model carries no systemic risk.

Hangs off: Article 53: GPAI model providers

Placed against the official source | examples

A police service fine-tunes an open general-purpose model on its own files from ongoing criminal investigations, so that detectives can see links between suspects and cases sooner. The fine-tuned model stays inside the service and is not made available to anyone else. The question is whether the service thereby becomes a provider of a general-purpose AI model itself, and so falls under Article 53.

Hangs off: Article 53: GPAI model providers

Placed against the official source | examples

Not every modification makes you a provider. The indicative threshold is a modification using more than a third of the original model's training compute, and your obligations are then limited to the modification.

Hangs off: Article 53: GPAI model providers

Placed against the official source | guidance

Documentation, information, copyright and transparency duties for providers of general-purpose AI models.

Placed against the official source | gpai

Public guide to GPAI model obligations and exceptions.

Hangs off: Article 53: GPAI model providers

Editorially reviewed | gpai, template