EU Seeks Assistance in Crafting Regulations for General-Purpose Artificial Intelligence Systems

The European Union is initiating a consultation phase regarding the regulations that will affect creators of general-purpose AI models (GPAIs), including industry leaders such as Anthropic, Google, Microsoft, and OpenAI, within its AI Act. This Act introduces a risk-based approach to the oversight of AI technologies. The anticipated Code of Practice aims to foster “trustworthy” GPAIs by laying out a roadmap for developers to adhere to mandatory legal standards.

The EU AI Act, having been ratified earlier within the year, is poised to be officially enforced starting August 1. However, it introduces a staggered timeline for compliance obligations, with Codes of Practice expected to be operational after nine months—aiming for an April 2025 timeline. This schedule provides the EU sufficient time to establish comprehensive guidance.

The Commission is calling upon feedback from GPAI developers who operate within the European Union, alongside input from corporations, societal organizations, rights advocates, and scholarly authorities.

“This consultation represents a crucial chance for diverse stakeholders to voice their perspectives on the inaugural Code of Practice, which will stipulate regulations for providers of general-purpose AI models,” the Commission stated. “Additionally, this initiative will guide subsequent endeavors by the AI Authority, especially regarding the format for the summarization of training data for general-purpose AI models and related directives.”

Structured as a questionnaire, the consultation delves into three major areas: transparency and copyright provisions for GPAIs, regulative measures addressing the systemic risk associated with GPAIs (as defined within the AI Act concerning computation thresholds), and oversight procedures for GPAI Codes of Practice.

An initial draft of the Code will be formulated “based on the responses and insights gained from the targeted inquiries,” the Commission has indicated.

Participants in this consultation are presented with an opportunity to shape the AI Office’s template designed for GPAI creators. This template must succinctly summarize the model training processes. The resultant template’s level of detail will be of interest to many.

Further details about the consultation are available here, with the submission deadline set for September 10, 2024, at 6 p.m. CET.

Moreover, the EU is soliciting participation interest for the Code’s drafting via online forums, split into four task groups. An iterative method will be employed to craft the guidance.

The AI Office is extending an invitation to “qualified general-purpose AI model developers, ancillary service providers, other industry associations, various stakeholder groups such as civil society or rights holder organizations, alongside academia and individual experts, to participate in the formulation of the Code of Practice.”

Expressions of interest to partake in the drafting process must be submitted by August 25, 2024, at 6 p.m. CET.

Furthermore, GPAI creators will have the opportunity to partake in workshops alongside the plenary meeting leaders. These sessions, according to the AI Office, aim to “enrich each drafting phase with insights, in addition to plenary engagements.”

“Efforts to maintain transparency in these dialogues will be made, for instance, through the preparation of meeting records made accessible to all plenary participants,” it was mentioned.

The AI Office will select chairpersons and vice-chairs, and is currently accepting applications from “interested independent specialists” for these pivotal roles.

Following concerns that civic organizations might be sidelined in the drafting process, the call for participation and consultation on the Code comes. Earlier, Euractiv reported about apprehensions related to the Commission’s reliance on consultancy firms for drafting the Code, raising alarms over a potentially biased process that may benefit AI majors.

To quell such apprehensions, the Commission has emphasized inclusivity. “We encourage all interested entities to partake,” it asserted on Tuesday. “The AI Office looks forward to submissions from a diverse array of stakeholders, including scholars, independent experts, industry players such as general-purpose AI model creators or downstream integrators, non-governmental organizations, rights owners, and governmental agencies.”