Background

The European Commission invited stakeholders to contribute to a Code of Practice for general-purpose artificial intelligence, meaning AI models intended for broad use. The initiative formed part of the continuing work connected with the EU AI Act. Its purpose was to develop practical expectations before the code was completed, helping providers and other actors understand the governance and documentation issues they would need to address.

Transparency and copyright

The consultation covered transparency around models and a copyright policy. The proposed code was therefore expected to address how providers describe their models and how they handle copyrighted material used during development. These questions connect technical choices, documentation and internal processes. The consultation did not itself establish a new binding standard, and its publication date did not automatically impose a complete set of operational duties.

Systemic risk

A second central theme was the identification, assessment and mitigation of systemic risk. The work was intended to cover risk assessments, mitigating measures and internal governance. For providers, this means security is not limited to testing a finished model. They also need the ability to explain which risks were considered, which controls were selected and how ongoing oversight is organised.

Training content

The Commission also included a template for a summary of content used to train models. A common template could support more consistent information about training material and make disclosures easier to compare. It remained, however, an element of the consultation and code-development process. It was not an independent requirement that automatically applied to every organisation when the consultation opened.

What the announcement meant

The announcement therefore marked a procedural beginning rather than the adoption of a finished rule. The Code of Practice was under development and should not be described as final binding guidance. Organisations could nevertheless use its topics as a preparatory checklist covering transparency, copyright, systemic-risk management and internal controls while the European process continued.

Norway and EU operations

For Norwegian model providers, the consultation was not Norwegian law and was not a final code. Providers supplying models into the EU needed to assess whether the AI Act’s territorial scope could become relevant to their activities. Domestic Norwegian application awaited EEA incorporation and Norwegian implementation. The European preparation process, later EU obligations and rules actually in force in Norway must therefore be kept distinct.

Sources

European Commission: “Consultation on the GPAI Code of Practice,” 30 July 2024.

EU AI Office: “Have your say on trustworthy general-purpose AI,” 30 July 2024.

For discussion

Should model providers begin documenting transparency, copyright and systemic-risk controls against the code’s themes even though the code was not final or binding?