The European Union has kicked off a consultation on rules that will apply to providers of general-purpose AI models (GPAIs) — such as Anthropic, Google, Microsoft and OpenAI — under the bloc’s AI Act, its risk-based framework for regulating applications of artificial intelligence. Lawmakers want the Code of Practice to help ensure “trustworthy” GPAIs by providing developers with guidance on how to comply with their legal obligations.
The EU AI Act was adopted earlier this year and will come into force imminently, on August 1. But it has a phased implementation for compliance deadlines, and Codes of Practice are due to apply after nine months — so April 2025. That gives the bloc time to draw up the guidance.
The Commission is inviting responses to the consultation from GPAI providers who have operations in the EU, as well as from businesses, civil society representatives, rights holders, and academic experts.
“The consultation is an opportunity for all stakeholders to have their say on the topics covered by the first Code of Practice, which will detail out rules for general-purpose AI model providers,” the Commission wrote. “The consultation will also inform related work by the AI Office, in particular on the template for the summary of the content used for the training of the general-purpose AI models and the accompanying guidance.”
The consultation is a questionnaire divided into three sections. One covers transparency and copyright-related provisions for GPAIs; the second is concerned with rules on risk taxonomy, assessment and mitigation for GPAIs with so called systemic risk (defined in the AI Act as models trained above a certain compute threshold); and the third section deals with the reviewing and monitoring of Codes of Practices for GPAIs.
The Commission said an initial draft Code will be developed “based on the submissions and answers to the targeted questions.”
Those responding to the consultation have the chance to influence the shape of the template the AI Office will provide to GPAI providers so they can fulfill a legal requirement to provide a summary of model training content. It will be interesting to watch how detailed that template ends up being.
More information on the consultation can be found here. The deadline for submissions is September 10, 2024, at 6 p.m. CET.
The EU is also calling for an expression of interest to participate in drawing up the Code via virtual meetings divided into four working groups. An iterative drafting process will be used to develop the guidance.
The AI Office is inviting “eligible general-purpose AI model providers, downstream providers and other industry organisations, other stakeholder organisations such as civil society organisations or rightsholders organisations, as well as academia and other independent experts to express their interest to participate in the drawing-up of the Code of Practice.”
The deadline for submitting an expression of interest to participate in the drafting is August 25, 2024, at 6 p.m. CET.
In addition, GPAI providers will get the chance to attend workshops with the plenary meeting chairs and vice chairs. These workshops, the AI Office said, are intended to “contribute to informing each iterative drafting round, in addition to their Plenary participation.”
“The AI Office will ensure transparency into these discussions, such as by drawing-up meeting minutes and making these available to all Plenary participants,” it noted.
The AI Office itself will appoint meeting chairs and vice chairs. It is taking applications from “interested independent experts” for these key steering roles.
The call and the consultation on the Code follow concerns that civil society organizations might be excluded from the drafting process. Earlier this month, Euractiv reported that the Commission was intending to rely on consultancy firms to draft the Code, leading to concerns of a skewed process that could favor AI giants.
The Commission sounds keen to dispel any such concerns. “All interested parties are encouraged to participate,” it wrote on Tuesday. “The AI Office invites submissions from a broad range of stakeholders, including academia, independent experts, industry representatives such as general-purpose AI model providers or downstream providers integrating the model into their AI systems, civil society organisations, rightsholders, and public authorities.”
