EU AI Act GPAI Obligations Take Effect: Transparency, Copyright and EU Representative Rules Bite

On 2 August 2025, the EU AI Act’s obligations for general-purpose AI models took effect. The date marks the EU’s shift from drafting rules to enforcing them.

Three new obligations after the effective date

Transparency. All general-purpose model providers must maintain and publish a summary of their training data, describing the model’s capabilities and limitations and the categories and sources of the training data.

Copyright. Providers must adopt and implement a copyright compliance policy, identify rights-holder reservations of rights and avoid generating infringing output.

Representation. Providers established outside the EU but placing general-purpose models on the EU market must appoint an authorised representative within the EU to act as the accountable interface with regulators.

The extra burden on systemic-risk models

For the most advanced models whose computing scale or impact crosses the threshold, the law adds heavier obligations: model evaluation, adversarial testing, serious-incident reporting and cybersecurity protection.

This tiered structure reflects the AI Act’s consistent logic: general obligations apply to all models, while enhanced obligations attach only to systemic-risk models. It avoids a one-size-fits-all burden on smaller providers and concentrates regulatory resources on the few actors posing the greatest risk.

Division of labour with the code of practice

Closely tied to the effective date is the General-Purpose AI Code of Practice published in July. See the EU GPAI code of practice.

The code translates those obligations into demonstrable, specific actions. A provider that follows it is better placed in a regulatory assessment, but the code is not itself the source of the obligations - that remains the text of the AI Act.

What it means for Chinese companies going global

First, assess whether you are placing a model on the EU market. If the service can be accessed by EU users, it generally falls within scope. See cross-border regulatory exposure for AI exporters.

Second, arrange the EU representative and compliance documentation. The appointment and the training-data summary need to be prepared in advance.

Third, check your copyright policy and training-data sources. Copyright is the area most likely to trigger disputes in EU enforcement. See GEMA v OpenAI.

A timetable still in flux

Notably, the EU is simultaneously pursuing a simplification and delay agenda, and the timeline for high-risk systems is under repeated discussion. See the European Parliament’s position on the AI Act omnibus.

This means companies cannot plan solely against the original timetable; they must track legislative developments and keep their compliance plans adjustable.

The entry into force of general-purpose model obligations is a watershed for EU AI governance: from this day, the model layer is no longer an avoidable compliance zone. See the EU AI Act transparency rules.

This article is based on public information on the effective arrangements for the EU AI Act’s general-purpose AI model obligations and is for reference only; it does not constitute legal advice.