The AI Office of the European Commission released last Tuesday the third draft of the Code of Good Practices designed to help providers of General Purpose AI (GPAI) models comply with the requirements induced by the AI Act, particularly regarding transparency, copyright, and risk management. This new version provides, according to the independent experts tasked with drafting the Code, a more streamlined structure as well as more nuanced commitments and measures.
While the AI Act came into effect in August 2024, its rules concerning GPAI such as OpenAI's GPT-4 or Google DeepMind's Gemini will come into application starting next August. Initiated by the European Commission, the Code of Good Practices aims to provide recommendations and guidelines to ensure responsible and transparent use of AI.
GPAI providers as well as civil society organizations, academics, and experts can participate in one of the working groups of the virtual plenary sessions organized by the AI Office for its drafting. The launch plenary session took place last September, followed by 2 online meetings that resulted in the first two drafts of the Code.
This new version, based on feedback received on the second version published last December 19, presents more detailed measures to implement each commitment. Two key commitments concern transparency and copyright. They apply to all general-purpose AI models, with a transparency exemption for open-source models. For copyright, the Code preserves the principles established in the previous version while making them more readable, with a less incisive tone. The measure concerning complaint filing allows signatories to refuse to follow up on a complaint "when the rights holders' complaints are manifestly unfounded or excessive, particularly due to their repetitive nature".
The other sixteen target providers whose models are classified as systemic risk, emphasizing safety and security. The measures can be found on the dedicated website (unofficial).
The initial reactions to this publication are very mixed. Several NGOs and digital rights advocates criticize the Commission for yielding to the pressure from big tech companies, arguing that the adjustments made weaken the regulation initially envisioned, while these companies consider the Code remains too restrictive and could stifle innovation.
Experts admit that this 3rd draft "still does not contain the level of clarity and coherence expected". Stakeholder comments expected by March 30 will allow them to refine it to present the final version scheduled for next May.