Firm News

Presentation of the First Draft of the Code of Practice for General-Purpose Artificial Intelligence or General-Purpose AI Code of Practice (“GPAICP”)

Carlos Lalanda ·

  1. CONTEXT
  • Article 56 of the AI Regulation introduces codes of practice as a tool provided by the European AI Office to contribute to the proper application of the Regulation, specifically to the obligations laid down in Articles 53 and 55 for providers of general-purpose AI models.
  • On 30 September the European AI Office launched the process of drafting the GPAICP. This is interactive and involves thousands of stakeholders, including academia, independent experts in intellectual property and personal data protection, providers of general-purpose AI models and providers of systems incorporating them, rights holders and public authorities. In addition to representatives of EU Member States and international observers.
  • The GPAICP is drawn up by an independent group of experts, the chairs and vice-chairs of 4 working groups (The main topics):
    • Transparency and copyright-related rules.
    • Identification and assessment measures for systemic risks.
    • Systemic risk mitigation measures.
    • Internal risk management and governance for providers of general-purpose AI models.
  • The process consists of 4 drafting phases, and the draft presented on 14 November by the AI Office and the European Commission is the outcome of the first of these. Its presentation was followed by a period of discussion of this 1st draft, which has just concluded, as the period for submitting feedback ended on 28 November.
  • The Code has neither binding force nor interpretative authority in relation to the Regulation. However, it is intended to serve as guidance for model providers while standardisation rules are being prepared. This is intended to prevent the legal uncertainty caused by the ambiguity of the Regulation from hindering innovation.
  • The intention is for the final document to be ready before 1 May 2025, since the obligations under the AI Regulation for GPAI models enter into force in August 2025.
  1. SUMMARY

Objectives of the General-Purpose AI Code of Practice

  • Achieve realistic applicability of the AI Regulation, facilitating innovation while respecting the rules and fundamental rights at stake.
  • Detail the obligations laid down in the AI Regulation. Translate general principles into measures that can be applied in practice.
  • Provide legal certainty and clarity to European market participants to encourage investment.

Content of the First Draft.

It is divided into the various obligations, for which it sets out measures and their corresponding sub-measures and KPIs.

  • Principles and objectives of the measures, sub-measures and KPIs:
    • Proportionality of the measures and KPIs to the risks and the size of the provider.
    • Simplified compliance for SMEs.
    • Exceptions for open-source models.
    • Support for the AI safety ecosystem.
    • Alignment with EU rights and values and the AI Regulation.
    • Alignment with the international approach.
  • Obligations for providers of general-purpose AI models:
    • Includes detailed documentation requirements and proposes information sharing with the authorities.
    • Compliance with intellectual property rules.
    • Taxonomy of systemic risks.
  • Obligations for providers of general-purpose AI models that may pose systemic risks:
    • Risk assessment.
    • Technical mitigation measures.
    • Governance mitigation measures
  • Open questions on key topics awaiting further development.

More publications