Summary No, the proposed Cloud and AI Development Act (CADA) does not require frontier AI projects to be open source. Article 8 of the proposal sets specific criteria for designating a "frontier AI priority project," focusing on the project's pioneering nature, multi-state participation, and the pooling of computing resources, but it contains no mandate regarding software licensing or code transparency. While CADA promotes open source generally for public sector bodies under Articles 41–44, this policy preference applies to public procurement and software reuse, not to the fundamental research and development of frontier AI models by consortia. Consequently, a project can be designated as a priority and receive Union support while remaining proprietary.

Detail

As proposed, the Cloud and AI Development Act (CADA) establishes a clear distinction between the strategic support of frontier AI development and the regulatory framework governing public sector software adoption. For technical leaders, researchers, and consortium members evaluating the legislation, it is critical to understand that the designation of a "frontier AI priority project" under Article 8 is a mechanism for securing Union support and access to computing resources, not a condition for open-source compliance.

Article 8: The Specific Criteria for Designation

Article 8 of the CADA proposal outlines the exclusive conditions under which the European Commission may recognize a project as a "frontier AI priority project." The text states that the Commission may recognize projects selected through open calls for expression of interest that support "grand challenge 3" set out in Annex I, provided they fulfill three cumulative criteria:

  1. Pioneering Nature: The project must be "a pioneering project, focused on the support and scaling-up of frontier AI technologies."
  2. Multi-State Participation: The project must be undertaken by a European digital infrastructure consortium (established pursuant to Decision (EU) 2022/2481) or another legal entity eligible for funding under Union law, and it must "involve the participation of at least three Member States."
  3. Pooled Resources: The participating Member States must "pool computing time and other relevant resources to support the implementation of the designated project."

Notably, none of these criteria reference software licensing, open-source standards, or the public availability of model weights, training data, or source code. The legislative focus is entirely on the collaborative, cross-border nature of the effort and the strategic importance of scaling up frontier AI capabilities within the Union. The absence of any reference to open source in Article 8 confirms that licensing status is not a gatekeeper for designation.

Article 9: Computing Support Without Licensing Conditions

Complementing Article 8, Article 9 mandates that the Union and Member States ensure sufficient AI computing resources are allocated to support these designated frontier AI priority projects. It further requires the Union to "match, on a proportional basis and within the limits of available European high-performance computing ('EuroHPC') capacity, the AI computing resources contributed or committed by the Member States to the designated frontier AI priority projects."

Like Article 8, Article 9 is silent on intellectual property rights or open-source requirements. Its sole objective is to ensure that strategically important AI projects have access to the necessary computational power to bridge the capacity gap. The provision does not condition this access on the release of code or models.

The Separate Open Source Framework (Articles 41–44)

The confusion regarding an open-source requirement for frontier AI likely stems from CADA's broader provisions on open source, found in Articles 41 through 44 under Title IV (Autonomy). These articles establish a strong preference for open-source solutions, but their scope is strictly limited to the public sector:

  • Article 41 encourages Union entities and public sector bodies to "use and facilitate the reuse of open standards and components released under an open source licence when building their cloud and AI ecosystem or stack."
  • Article 42 requires that when Union entities or public sector bodies make software available for reuse under an open-source licence, they must do so using a catalogue connected to the EU OSS Catalogue.
  • Article 43 establishes the EU Open Source Solutions Catalogue as a centralized resource for public sector software.
  • Article 44 creates a network of Open Source Programme Offices (OSPOs) to facilitate cooperation among public entities.

However, these provisions target public sector procurement and software reuse, not the fundamental research and development of frontier AI models by consortia or private entities. A frontier AI priority project is a research and innovation initiative aimed at scaling technology; it is not a public sector procurement of a cloud service or a public body releasing internally developed administrative software. Therefore, the open-source preferences in Articles 41–44 do not apply to the core development activities of frontier AI priority projects.

Scope Boundary: Research vs. Procurement

It is essential to distinguish between the Cloud and AI Leadership Initiatives (Title II of CADA), which support R&D and capacity building, and the Autonomy and Adoption measures (Title IV), which govern procurement and sovereignty. Frontier AI priority projects fall squarely under the former. Their purpose, as stated in the explanatory memorandum, is to "bridge the gap between the Union's advanced research and innovation capabilities and their sustainable exploitation." The EU's goal here is to ensure European entities have the capability to develop and deploy frontier AI, regardless of whether the resulting models are proprietary or open source.

In contrast, the open-source provisions in Title IV are designed to reduce vendor lock-in for public administrations and promote interoperability in public sector IT stacks. These are distinct policy objectives with separate legal bases and scopes of application. The proposal does not conflate the two; a project can be a "frontier AI priority project" (Title II) without being subject to the public sector open-source mandates (Title IV).

What this means for you

For CTOs, architects, researchers, and SMEs involved in or evaluating frontier AI projects, the absence of an open-source requirement in Article 8 offers significant strategic flexibility:

  1. IP Strategy Preservation: You can pursue proprietary models and closed-source architectures for your frontier AI initiatives without losing eligibility for Union support or computing resources. The proposal does not force you to release your competitive advantage (model weights, training data, or algorithms) as a condition for being designated a priority project.
  2. Focus on Collaboration, Not Code: To qualify as a frontier AI priority project, your primary focus should be on securing partnerships with at least three Member States and demonstrating a pioneering approach to scaling frontier AI. Ensure your consortium structure and resource-pooling mechanisms align with Article 8(b) and 8(c).
  3. Public Sector Clients: If your frontier AI project eventually leads to a product or service procured by a public sector body, the open-source preferences in Article 41 may become relevant at the procurement stage. However, this is a downstream commercial consideration, not a condition for the initial R&D designation.
  4. Compliance Monitoring: Monitor the development of Annex I (Grand Challenges) and any delegated acts amending it, as the specific technical definitions of "pioneering" may evolve. However, based on the current text of Article 8, open-source status is not a criterion.

Common misconceptions

  • "CADA forces all EU AI research to be open source." This is incorrect. CADA promotes open source for public sector software reuse and procurement (Articles 41–44) but does not impose open-source mandates on private R&D or frontier AI priority projects.
  • "To get EU compute time, you must open-source your model." False. Article 9 guarantees compute time for designated frontier AI priority projects based on the criteria in Article 8, which are purely about pioneering nature, multi-state involvement, and resource pooling.
  • "Frontier AI projects are subject to the same rules as public cloud procurement." No. Frontier AI projects are R&D initiatives supported by the Cloud and AI Leadership Initiatives. Public cloud procurement is governed by the sovereignty framework (Articles 16–33), which includes open-source preferences but also allows for proprietary solutions if they meet sovereignty assurance levels.

Official sources

Related

This is general information about a draft EU regulation, not legal advice.