GPAI authorised representative: check the model role and open-source exception
The acquired AI Act Article 54 addresses third-country providers placing general-purpose AI models on the Union market. It specifies a written mandate and includes an open-source exception with conditions and a systemic-risk limitation. Confirm the model-provider role, release facts and applicable version/dates before buying representation.
Evidence retrieved 2026-10-06. Source versions and topic-specific limits are listed below.
Sourced criteria · EU
What changes the service scope?
Decision or task
What the source describes
What to prepare
Is this the GPAI model provider?
Article 54(1) addresses providers of general-purpose AI models established in third countries, before Union market placement. [1]
Distinguish the model provider from a downstream application/deployer and record placement and establishment facts.
Does the open-source exception apply?
Article 54(6) requires the described licence/access conditions and public parameters including weights, architecture and usage information; the exception does not apply to models presenting systemic risks. [1]
Compare every stated condition; do not infer the exception from an “open” marketing label alone.
What must the mandate enable?
Article 54(3)–(4) includes documentation/obligation checks, document retention, information provision and authority cooperation. [1]
Agree document access, covered model versions, reasoned-request handling and escalation.
Which timing and legacy facts matter?
Articles 111 and 113 contain separate application and existing-model provisions in this dated version. [2][3]
Read these provisions against the model’s first-placement date and current authentic amending acts; avoid copying an old generic AI deadline.
This dated consolidated text is a reference; authentic amending acts and deciding model/systemic-risk facts need review. This page does not apply GPAI representation to every AI application.
A scoped evidence and handoff plan
Use this checklist to gather your business or product details before speaking with a specialist. The items below explain what to record and suggest useful supporting documents. You can add your own answers in the editable project brief.
Is this the GPAI model provider?
Distinguish the model provider from a downstream application/deployer and record placement and establishment facts.
Useful evidence: Model/version, provider/entity, distribution and first-placement records.
Does the open-source exception apply?
Compare every stated condition; do not infer the exception from an “open” marketing label alone.
Useful evidence: Licence, public weights/architecture/usage links and systemic-risk assessment inputs.
What must the mandate enable?
Agree document access, covered model versions, reasoned-request handling and escalation.
Useful evidence: Mandate, technical-document index and authority/contact workflow.
Which timing and legacy facts matter?
Read these provisions against the model’s first-placement date and current authentic amending acts; avoid copying an old generic AI deadline.
Useful evidence: Placement-date evidence, version history and a reviewed applicability-date record.
Work packages and dependencies
Conditional: Model-provider, exception and mandate assessment — For this work package, agree the supported criteria, evidence access, covered entities/products and unresolved facts in the table above. Additional services require their own justified scope.
Questions for providers
Which exact actor, product or processing facts support the quoted scope, and what is still unresolved?
How will the listed records reach the responsible people, and who owns each change or authority request?
Which tasks and entities are excluded from the agreement, and which additional services need a separate assessment?
Sources and data dates
Read the official document in context. The audit details identify the precise locators and preserved versions used for this page.
