The European Commission has initiated a targeted consultation as part of its efforts to prepare guidelines that clarify the obligations of providers of general-purpose artificial intelligence (AI) models under the EU's AI Act. This consultation, led by the Commission's AI Office, aims to gather input from a wide array of stakeholders including industry representatives, civil society groups, academic experts, independent authorities, and public officials. The overarching goal is to provide clear and practical guidance on key areas such as defining what constitutes a "general-purpose AI model," the regulatory process for introducing these models into the European Union market, estimating the computational resources necessary for training or modifying such models, and outlining mechanisms for supervision and enforcement of relevant rules.

At the heart of the consultation is the challenge of defining a "general-purpose AI model." The Commission's preliminary view frames such a model as one exhibiting "significant generality" capable of effectively performing a diverse set of tasks. An important metric under consideration is the amount of training compute, measured in floating-point operations (FLOP). Currently, the European Commission considers training compute a practical though imperfect proxy for assessing a model's generality and capabilities. Specifically, a threshold has been proposed whereby models capable of generating text and/or images with training compute exceeding 10^22 FLOP would typically be categorised as general-purpose AI. However, this classification is subject to rebuttal if evidence demonstrates that a model's functionality is narrower in scope.

The Commission plans to release these guidelines alongside a General-Purpose AI Code of Practice, which is expected to establish commitments for compliance related to documentation standards, adherence to copyright laws, and management of systemic risks where applicable. Both documents are targeted for publication by May or June 2025.

The consultation process remains open for contributions until 12:00 noon Central European Time on Thursday 22 May 2025. The AI Office has encouraged all interested parties to participate by submitting their feedback during this period.

The information presented here draws on a report published by Mondaq and aims to provide an informative outline of the European Commission's ongoing regulation efforts. For those impacted by the evolving AI regulatory landscape, seeking specialist advice tailored to individual circumstances is recommended.

Source: Noah Wire Services