📢EU PUBLISHES FIRST CODE OF PRACTICE FOR GENERAL-PURPOSE AI MODELS: A MILESTONE FOR RESPONSIBLE AI GOVERNANCE

On 10 July 2025, the European Commission officially unveiled the Code of Practice for General-Purpose AI Models (GPAI Code), marking a critical step toward ensuring responsible AI innovation across the EU.

Designed through a multi-stakeholder process and built as a voluntary tool, the Code helps AI model providers demonstrate compliance with the transparency, safety, and copyright obligations under the AI Act, particularly Articles 53 and 55.

Why this Code matters

While general-purpose AI (GPAI) models such as large language models or multimodal systems drive rapid innovation, they also pose systemic risks ranging from misinformation to intellectual property infringements. The GPAI Code is the EU’s proactive response, developed ahead of the binding AI Act obligations entering into force in August 2025 (for Article 53) and August 2026 (for Article 55).

Structure and Scope

The Code is structured into three dedicated chapters, each addressing a core compliance area:

1. Transparency (Article 53(1)(a)-(b))

This chapter obliges GPAI providers to maintain robust Model Documentation, using a standardised Model Documentation Form. It ensures downstream providers and authorities can understand a model’s architecture, capabilities, design rationale, and limitations. It also includes provisions to keep prior versions archived for 10 years and to make contact channels available for data access or clarification requests.

2. Copyright (Article 53(1)(c))

Providers must implement a copyright policy aligned with Directive (EU) 2019/790, focusing on:

• Avoiding the use of unlawfully accessible data during training

• Ho nouring machine-readable rights reservations using standards like robots.txt

• Implementing technical safeguards against copyright-infringing outputs

• Setting up complaint mechanisms for rightsholders

  
  

This chapter clarifies that while adherence to the Code helps demonstrate compliance, it does not constitute legal proof under Union Copyright Law.

3. Safety and Security (Article 55 – systemic risk models)

This section applies only to advanced GPAI models considered to pose systemic risks. It introduces a detailed Safety and Security Framework and a lifecycle-based approach to risk governance. Key commitments include:

• Conducting continuous model evaluations

• Defining risk thresholds and response plans

• Reporting serious incidents to the AI Office and National Competent Authorities

• Collaborating with independent evaluators and civil society

• Implementing state-of-the-art safety and cybersecurity protocols

  
  

Notably, small and medium-sized providers benefit from proportionality clauses, allowing simplified reporting and compliance pathways.

Evaluation Phase and Commission Guidelines

The European Commission and Member States will now assess the adequacy of the Code in the coming weeks. Furthermore, the Code will be complemented by Commission guidelines later in July 2025 to clarify key concepts and further support implementation.

Stakeholders across the AI value chain, including model providers, regulators, researchers and rightsholders, are encouraged to engage with the Code. Their participation supports the development of a trustworthy AI ecosystem that promotes innovation while ensuring ethical and legal safeguards.

Implications for the AI Ecosystem

The GPAI Code of Practice represents a first-of-its-kind regulatory innovation, blending soft law and hard law to guide the AI industry toward compliance and accountability. It sends a clear message: the EU remains committed to advancing AI responsibly, with no grace periods or regulatory pauses.

As the implementation of the AI Act accelerates, voluntary adherence to the Code may offer a competitive advantage, both in demonstrating regulatory readiness and in building public trust.

For tailored legal advice on compliance with the EU AI Act and the General-Purpose AI Code of Practice, please contact our team at info@pelaghiaslaw.com. We are available to support clients navigating the evolving European AI regulatory framework.