EU AI Office Convenes Second GPAI Code of Practice Meeting Focused on Copyright Chapter

The European Commission announced on April 8, 2026, that the EU AI Office had convened the second meeting of the Signatory Taskforce under the General-Purpose Artificial Intelligence (GPAI) Code of Practice. Representatives from major technology companies including Amazon, Google, and Microsoft participated in the session, which centered on the Copyright Chapter of the GPAI Code of Practice — the provisions governing how AI model providers must mitigate copyright-infringing outputs and establish complaint mechanisms for rights holders.

Two substantive themes dominated the discussion: technical approaches for mitigating copyright-infringing outputs from generative AI systems, and the design and functioning of contact points through which rights holders can lodge copyright complaints. Participants shared implementation experiences and discussed attribution algorithms alongside lifecycle-based approaches that address copyright risk at the point of training data selection rather than only at the output stage.

The GPAI Code of Practice and Its Legal Foundation

The GPAI Code of Practice derives its authority from the EU AI Act (Regulation (EU) 2024/1689), which entered into force in August 2024 and applies to providers of general-purpose AI models. The Code is intended to operationalize the obligations set out in Articles 53 and 55 of the AI Act, including transparency requirements, compliance with copyright law, and testing and evaluation requirements. For GPAI model providers designated as posing systemic risk, compliance with the Code serves as a presumption of conformity with the Act’s more stringent obligations.

The Copyright Chapter specifically addresses the intersection of AI training practices and the rights established under Directive 2019/790 on copyright in the Digital Single Market. Article 4 of that Directive grants rights holders the ability to opt out of text and data mining (TDM) uses of their works — a right that AI providers must honor when assembling training datasets for deployment within the EU. The Copyright Chapter of the GPAI Code translates this legal obligation into practical technical and procedural requirements.

Attribution Algorithms and Lifecycle Approaches

The attribution algorithms discussed during the meeting refer to technical mechanisms that track or signal the provenance of content generated by AI models — enabling rights holders to identify when their works may have influenced a model’s outputs, and providing a basis for exercising opt-out rights or pursuing compensation claims. The technical implementation of attribution varies considerably across companies, and no standardized approach has yet emerged within the industry.

The lifecycle approach represents a more systemic alternative: rather than addressing copyright risk at the output stage alone, it involves continuous evaluation and mitigation of copyright exposure throughout the model development pipeline — from initial data selection and curation through fine-tuning and deployment. The European Commission has signaled a preference for input-level risk controls of this kind, viewing them as more structurally sound than post-hoc filtering of potentially infringing outputs.

Complaint Contact Points: Design Challenges

The second major agenda item — the design and functioning of contact points for rights holder complaints — poses its own difficulties. A functional complaint mechanism must be accessible to rights holders across multiple jurisdictions, capable of handling complaints under different national copyright frameworks, and responsive enough to be practically useful. Rights holder representatives have expressed concern that complaint mechanisms proposed by AI companies may prove difficult to navigate and slow to produce results in practice.

Participants at the second meeting shared details of their current contact point designs and operational experience, with the goal of informing the ongoing standardization work under the taskforce.

Significance for Global AI Copyright Policy

The EU’s approach to AI and copyright — anchored in an opt-out right for TDM and operationalized through the GPAI Code — stands in contrast to frameworks elsewhere. Japan’s Copyright Act Article 30-4 takes a more permissive approach, broadly allowing AI training uses of copyrighted works without requiring opt-out compliance. The United States has no equivalent opt-out framework. As the EU’s implementation experience accumulates, the practical differences between these regulatory models will become clearer and may inform future policy debates in other jurisdictions.

The development of industry-wide standards for attribution and lifecycle copyright management under the GPAI Code may also have implications for AI patent practice, as the technical measures themselves may become the subject of patent filings.

This development was reported by IPWatchdog on April 10, 2026.