Digital for Planet
CONTACT US

JOIN OUR TEAM →

Chapter V of the AI Act applies as of 2 August 2025: European Commission Guidelines on the Scope of Obligations for General-Purpose AI Models

The European Commission’s Guidelines on the Scope of the Obligations for General-Purpose AI Models (‘Guidelines’) were published on 18 July 2025, ahead of today’s entry into application of Chapter V of the AI Act (Regulation 2024/1689). The Guidelines, now approved, aim to clarify how the Commission interprets and will enforce the AI Act’s provisions on general-purpose AI (GPAI) models—including those with systemic risk. They lay out compliance expectations for AI developers and deployers, especially those handling high-impact, multi-purpose models.

The Guidelines cover four main areas:

  • definitions and classification of GPAI models;
  • obligations for providers;
  • exemptions for open-source models; and
  • enforcement by the AI Office.

The general context of the Guidelines aligns with Digital for Planet’s (D4P) mission: promoting sustainable, inclusive, zero‑pollution digital space in Europe, including, among others, efficient and ethical AI. We have unpacked the Guidelines and focused on three points that are the most prominent for D4P mission:

  • the use of generative AI in research (including for writing proposals under Horizon Europe and other European R&I funding schemes);
  • open-source model governance;
  • energy efficiency.

Key elements

1. Generative AI in research & proposal writing
 
The Guidelines explicitly exclude AI models used purely for research, development, or prototyping from the regulatory obligations—until they are placed on the market. This carve-out protects academic freedom and exploratory research. Moreover, general-purpose generative models (such as those used for language, text-to-image, or code generation) are recognised as critical research infrastructure. In the context of Horizon Europe Pillar II calls, these models can be instrumental in proposal drafting, literature review, scenario simulation, and such like—of course, subject to a compulsory, systematic, and thorough human oversight. However, once such tools are embedded in commercial R&I outputs or services, providers must comply with transparency, documentation, and copyright obligations under Article 53 of the AI Act. The guidelines support the use of generative AI in shaping competitive, compliant proposals—so long as AI usage is transparently disclosed and responsibly governed.

 

 

2. Open-source model governance
 
A pivotal element of the guidelines is the structured exemption for open-source AI models, provided they meet stringent conditions: (i) non-monetised access, (ii) publicly available weights and architecture, and (iii) truly permissive licences (excluding “non-commercial use only” clauses). These exemptions aim to protect grassroots innovation while maintaining EU standards of transparency and accountability. Crucially, even open-source models are not exempt if they are deemed to have systemic risk (i.e., high-impact capabilities). This balancing act between openness and safety offers a regulatory template for community-led AI development. In Horizon Europe projects, consortia leveraging open-source AI should ensure compliance with these exemption criteria to maintain eligibility and minimise administrative burdens.

 

 

3. Energy efficiency in AI model lifecycle
 
While not a primary headline in the document, energy efficiency is implicitly framed through the AI Act’s requirement to estimate and report training compute—a measure of energy-intensive computation quantified in FLOPs (floating-point operations). Models crossing 10²³ FLOP are presumed to be general-purpose AI models, while 10²⁵ FLOP triggers classification as having systemic risk. This regulatory attention to compute cost promotes design efficiency, energy transparency, and possibly future environmental sustainability benchmarks. Energy-efficient training strategies (e.g., model distillation, quantisation) could thus form a competitive advantage and align with
objectives in research proposals.

 

 

Shared vision

The Guidelines are of critical relevance for actors seeking Horizon Europe Pillar II funding, as they intersect with regulatory, research, and ethical innovation priorities. Hence, also for D4P, our partners, our clients, and our supporters.

D4P works on several Horizon Europe projects and proposals on efficient and ethical AI (see, e.g., CERTAIN, COP-PILOT, etc.). Therefore, we could not miss these Guidelines and AI Act Chater V application. D4P’s work across climate-neutral, inclusive, and responsible AI hinges on efficient, harmonised rules for AI, including GPAI. Connected to this, the new Guidelines’ regulatory foresights that we support, include, for example:

  • Streamlined obligations — because it will leverage the capacity in European research, including to address transparency, responsibility, and energy efficiency of AI, and broader inclusivity, protection of privacy and fundamental rights, and conservation of nature, climate, and biodiversity;
  • Enforcement — because it will likely further nurture innovation in sustainable AI-driven and co-created tools, and further digital tools and services.

 D4P invites stakeholders to join the conversation—visit our #DecodeGreen portal and subscribe to our newsletter for updates.

Share with Your Network

LinkedIn
Email