How does CADA interact with the AI Act for AI deployed by public administrations?

The proposed Cloud and AI Development Act (CADA) does not replace the EU AI Act's obligations for public administrations; rather, it layers mandatory sover

Summary The proposed Cloud and AI Development Act (CADA) does not replace the EU AI Act's obligations for public administrations; rather, it layers mandatory sovereignty and procurement rules on top of them. While the AI Act governs the safety, fundamental rights, and transparency of the AI systems themselves, CADA Articles 29 and 30 dictate where and how these systems must be hosted. As proposed, CADA requires public bodies to conduct specific risk assessments to determine the necessary "Union assurance level" for their cloud infrastructure. This creates a dual-compliance regime: the AI Act ensures the AI is trustworthy, while CADA ensures the cloud beneath it is sovereign and resilient against third-country interference.

Detail

For public-sector procurement officers, digital leaders, and legal counsel, the interaction between the proposed Cloud and AI Development Act (CADA) and the EU AI Act (Regulation (EU) 2024/1689) represents a fundamental shift from purely functional compliance to structural sovereignty. The AI Act establishes the baseline for the quality and risk management of AI systems, while CADA establishes the baseline for the infrastructure and supply chain resilience of the services hosting them.

The AI Act Baseline: Deployer Obligations

Under the AI Act, public administrations acting as "deployers" of high-risk AI systems (as defined in Annex III of the AI Act, such as systems for emergency services, justice, migration, or law enforcement) must adhere to strict operational obligations. These include:

Human Oversight: Ensuring natural persons can oversee the system's operation to prevent or mitigate risks (AI Act Article 26).
Fundamental Rights Impact Assessments: Conducting assessments prior to deployment to identify risks to health, safety, and fundamental rights (AI Act Article 27).
Monitoring and Logging: Continuously monitoring the system's operation and keeping logs of its functioning to ensure traceability (AI Act Article 26).

CADA explicitly respects these obligations. The explanatory memorandum of the CADA proposal notes that the AI Act "harmonises rules for AI systems and general-purpose AI models to be placed on the EU market" and "ensures a high level of protection of health, safety and fundamental rights." However, the memorandum explicitly states that the AI Act "does not cover aspects of sovereignty." This regulatory gap—specifically regarding where compute is located, who controls the provider, and whether foreign laws could compel access to the infrastructure—is what CADA is designed to fill.

CADA's Sovereignty Layer: Articles 29 and 30

CADA introduces a mandatory framework for the procurement of cloud computing services by public authorities, directly impacting how AI systems are hosted and operated. This is governed primarily by Article 29 and Article 30 of the proposed regulation.

Article 29: Mandatory Risk Assessments CADA obliges Member States and Union entities to carry out risk assessments to determine the appropriate level of sovereignty required for their cloud-based activities.

Scope: These assessments must identify public sector activities that contribute to the preservation of public order. This includes sectors falling under Annex I or II of the NIS2 Directive, as well as national security, internal security, external border management, defence, justice, and law enforcement (Article 29(1)).
Frequency: Assessments must be conducted by the date of entry into force plus one year, and thereafter every two years, or whenever necessary (Article 29(1)).
Criteria: The assessment must evaluate the sensitivity, criticality, and magnitude of the data processed, the risk of unlawful access by a third country or legal entity established in a third country, and the risk of service disruption (Article 29(2)).
Outcome: The risk assessment determines which "Union assurance level" (1, 2, 3, or 4) is appropriate for the activity. Union assurance levels 3 and 4 are designed for the most critical use cases, ensuring that data remains under effective EU supervision and is protected from extraterritorial access.

Article 30: Procurement Mandates Based on Risk The results of the Article 29 risk assessment directly dictate procurement behavior under Article 30.

Baseline Requirement: Union entities and public sector bodies whose activities have not been identified as contributing to the preservation of public order must use cloud computing services recognized as having at least Union assurance level 1 (Article 30(2)).
Enhanced Requirement: Contracting authorities whose activities have been identified as contributing to public order (e.g., defence, justice, critical infrastructure) must only procure cloud computing services recognized as having Union assurance level 2, 3, or 4 (Article 30(3)).
Derogations: Exceptional derogations are permitted only if no recognized service exists, if previous tenders failed, or if compliance would result in disproportionate cost (Article 30(4)).

This creates a direct link between the AI Act's classification of a system as "high-risk" (due to its impact on fundamental rights) and CADA's requirement for sovereign hosting. A high-risk AI system used in law enforcement, for example, must not only meet the AI Act's technical and governance standards but also run on infrastructure that meets CADA's highest sovereignty criteria.

Strategic Support: Leadership Initiative Objective 7

Beyond compliance, CADA actively supports public-sector AI deployment through the Cloud and AI Leadership Initiatives. Specifically, Article 4(2)(g) outlines "operational objective 7," which aims to increase the development and adoption of AI models and systems across the Union's public sectors. This includes accelerating technological development in critical public domains, promoting the sharing and reuse of training data across public services, and facilitating secure health data reuse for AI tools. This objective ensures that public administrations are not just passive regulators but active participants in building a sovereign AI ecosystem.

What this means for you

For public-sector procurement officers, the interaction between CADA and the AI Act requires a two-track compliance strategy for every AI project:

Conduct a Dual Assessment: Before procuring an AI service, you must perform the fundamental rights impact assessment required by the AI Act and the sovereignty risk assessment required by CADA Article 29. The CADA assessment will determine the minimum Union assurance level your cloud provider must hold.
Verify Provider Recognition: You can no longer rely solely on a vendor's AI safety certifications. You must verify that their cloud infrastructure is registered in the CADA central repository (Article 22) with the specific assurance level mandated by your risk assessment. For critical public order activities, standard commercial cloud offers (often Level 1 or unverified) will be non-compliant.
Plan for Migration: If your current AI systems are hosted on non-compliant infrastructure, CADA Article 29(6) provides a transition period of up to 12 months to migrate to a compliant provider, taking into account technical feasibility and data portability.
Leverage Open Source: CADA Article 41 encourages the use of open-source solutions to reduce vendor lock-in and enhance sovereignty. When deploying AI, consider open-source models and infrastructure to maintain greater control over the supply chain, aligning with both CADA's sovereignty goals and the AI Act's transparency requirements.

Common misconceptions

"CADA replaces the AI Act for public bodies." Incorrect. The AI Act remains the primary law for the safety, accuracy, and fundamental rights compliance of the AI system. CADA regulates the cloud infrastructure hosting that system. You must comply with both.
"All public AI deployments require Union assurance level 4." Incorrect. CADA is proportionate. Only activities identified in the Article 29 risk assessment as contributing to public order (e.g., defence, justice) require levels 2, 3, or 4. Other public services may only require level 1.
"Sovereignty is only about data location." Incorrect. CADA's assurance levels (Annex II) cover much more than geography. They include requirements for personnel citizenship (for levels 3 and 4), absence of third-country control, software supply chain transparency, and the right to audit source code.
"The AI Act already covers cloud security." Incorrect. The AI Act focuses on the robustness and cybersecurity of the AI model itself. CADA addresses the broader sovereignty risks, including the extraterritorial reach of third-country laws (such as the US CLOUD Act) that could compel cloud providers to hand over data or disrupt services, which the AI Act does not address.

Official sources

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