The European Commission has received the final version of the Code of Practice on General Purpose Artificial Intelligence (GPAI) models.
    The Code, which must now receive approval from the Commission and the EU27, provides clarifications on a series of rules envisaged in the AI ;;Act that will apply, starting August 2, to GPAI model providers, particularly those with systemic risks such as OpenAI's GPT-4, Google's Gemini, and xAI's Grok.
    The Code is composed of three chapters: transparency and copyright, which will apply to all GPAI models, and the assessment and mitigation of risks affecting a limited number of models, the most powerful and advanced ones.

This is a voluntary tool developed by 13 independent experts, with input from over a thousand stakeholders, including model providers, SMEs, academics, AI safety experts, rights holders, and civil society organizations.
    Signatories to the Code will benefit from a reduced administrative burden and greater legal certainty compared to providers demonstrating compliance in other ways.
    Implementation deadlines have been extended: if GPAI model providers do not fully implement their commitments immediately after signing the Code, the AI ;;Office, established within the Commission, will not consider them to be in default, but will instead consider them to have implemented the Code in good faith.
    The Code will be complemented by Commission guidelines, which will be published before the General Purpose AI obligations enter into force.

The guidelines will clarify who does and does not fall within the scope of the AI ;;Act's general-purpose AI regulations.
    The circulated drafts of the Code, the final draft of which was expected in May, had sparked protests from industry insiders, who denounced pressure from Big Tech to water down the text.
    Further complicating the situation are ongoing negotiations on tariffs, with Washington accusing Brussels of penalizing the American tech sector with its digital rules. A double game that the EU has, on the contrary, tried to keep separate.
    "The text is not perfect because pressure from large developers has been felt," admitted Democratic Party MEP Brando Benifei, rapporteur for the AI ;;Act, "but now, even more so, it must be rigorously applied, with precise controls and sanctions for non-compliance or unwillingness to comply with the obligations." "Today's publication of the final version of the Code of Conduct for General Purpose AI marks an important step towards making the most advanced AI models available in Europe not only innovative, but also safe and transparent. Designed in collaboration with AI stakeholders, the Code is aligned with their needs," said Henna Virkkunen, Commission Vice-President for Technological Sovereignty, Security and Democracy, inviting "all providers of general purpose AI models to join the Code.
    "This," she emphasized, "will provide them with a clear and collaborative path to compliance with European AI law."
   

ALL RIGHTS RESERVED © Copyright ANSA